diff --git a/build.toml b/build.toml index 039d94ea52a91291d7a594832225480183d4c734..70e3c9941fee495c024f1df2db0c9696e455216e 100644 --- a/build.toml +++ b/build.toml @@ -1,34 +1,16 @@ [general] name = "sage_attention" -universal = false -cuda-minver = "12.0" +backends = ["cuda"] [torch] src = [ - "torch-ext/torch_binding.cpp", - "torch-ext/torch_binding.h", + "torch-ext/torch_binding.cpp", + "torch-ext/torch_binding.h", ] -[kernel._qattn] -depends = ["torch"] +[kernel._qattn_sm89] backend = "cuda" -cuda-minver = "12.0" -cuda-capabilities = [ - "8.0", "8.9", "9.0a" -] -src = [ - "sage_attention/cp_async.cuh", - "sage_attention/dispatch_utils.h", - "sage_attention/math.cuh", - "sage_attention/mma.cuh", - "sage_attention/numeric_conversion.cuh", - "sage_attention/permuted_smem.cuh", - "sage_attention/reduction_utils.cuh", - "sage_attention/wgmma.cuh", - "sage_attention/utils.cuh", - "sage_attention/cuda_tensormap_shim.cuh", -] -cxx-flags = ["-g", "-O3", "-fopenmp", "-lgomp", "-std=c++17", "-DENABLE_BF16"] +cuda-capabilities = ["8.9"] cuda-flags = [ "-O3", "-std=c++17", @@ -37,42 +19,18 @@ cuda-flags = [ "--use_fast_math", "--threads=1", "-Xptxas=-v", - "-diag-suppress=174", # suppress the specific warning -] - -[kernel._qattn_sm80] -depends = ["torch"] -backend = "cuda" -cuda-minver = "12.0" -cuda-capabilities = [ - "8.0" -] -include = ["."] -src = [ - "sage_attention/qattn/qk_int_sv_f16_cuda_sm80.cu", - "sage_attention/qattn/attn_cuda_sm80.h", - "sage_attention/qattn/attn_utils.cuh" + "-diag-suppress=174", ] -cxx-flags = ["-g", "-O3", "-fopenmp", "-lgomp", "-std=c++17", "-DENABLE_BF16"] -cuda-flags = [ +cuda-minver = "12.6" +cxx-flags = [ + "-g", "-O3", + "-fopenmp", + "-lgomp", "-std=c++17", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "--use_fast_math", - "--threads=1", - "-Xptxas=-v", - "-diag-suppress=174", # suppress the specific warning + "-DENABLE_BF16", ] - - -[kernel._qattn_sm89] depends = ["torch"] -backend = "cuda" -cuda-minver = "12.0" -cuda-capabilities = [ - "8.9", -] include = ["."] src = [ "sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn_inst_buf.cu", @@ -84,9 +42,12 @@ src = [ "sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_fuse_v_scale_attn_inst_buf.cu", "sage_attention/qattn/attn_cuda_sm89.h", "sage_attention/qattn/qk_int_sv_f8_cuda_sm89.cuh", - "sage_attention/qattn/attn_utils.cuh" + "sage_attention/qattn/attn_utils.cuh", ] -cxx-flags = ["-g", "-O3", "-fopenmp", "-lgomp", "-std=c++17", "-DENABLE_BF16"] + +[kernel._qattn_sm90] +backend = "cuda" +cuda-capabilities = ["9.0a"] cuda-flags = [ "-O3", "-std=c++17", @@ -95,24 +56,32 @@ cuda-flags = [ "--use_fast_math", "--threads=1", "-Xptxas=-v", - "-diag-suppress=174", # suppress the specific warning + "-diag-suppress=174", ] - - -[kernel._qattn_sm90] -depends = ["torch"] -backend = "cuda" -cuda-minver = "12.0" -cuda-capabilities = [ - "9.0a", +cuda-minver = "12.6" +cxx-flags = [ + "-g", + "-O3", + "-fopenmp", + "-lgomp", + "-std=c++17", + "-DENABLE_BF16", ] +depends = ["torch"] include = ["."] src = [ "sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu", "sage_attention/qattn/attn_cuda_sm90.h", - "sage_attention/qattn/attn_utils.cuh" + "sage_attention/qattn/attn_utils.cuh", +] + +[kernel._qattn] +backend = "cuda" +cuda-capabilities = [ + "8.0", + "8.9", + "9.0a", ] -cxx-flags = ["-g", "-O3", "-fopenmp", "-lgomp", "-std=c++17", "-DENABLE_BF16"] cuda-flags = [ "-O3", "-std=c++17", @@ -121,22 +90,68 @@ cuda-flags = [ "--use_fast_math", "--threads=1", "-Xptxas=-v", - "-diag-suppress=174", # suppress the specific warning + "-diag-suppress=174", +] +cuda-minver = "12.6" +cxx-flags = [ + "-g", + "-O3", + "-fopenmp", + "-lgomp", + "-std=c++17", + "-DENABLE_BF16", ] - -[kernel._fused] depends = ["torch"] +src = [ + "sage_attention/cp_async.cuh", + "sage_attention/dispatch_utils.h", + "sage_attention/math.cuh", + "sage_attention/mma.cuh", + "sage_attention/numeric_conversion.cuh", + "sage_attention/permuted_smem.cuh", + "sage_attention/reduction_utils.cuh", + "sage_attention/wgmma.cuh", + "sage_attention/utils.cuh", + "sage_attention/cuda_tensormap_shim.cuh", +] + +[kernel._qattn_sm80] backend = "cuda" -cuda-minver = "12.0" -cuda-capabilities = [ - "8.0", "8.9", "9.0a", +cuda-capabilities = ["8.0"] +cuda-flags = [ + "-O3", + "-std=c++17", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "--use_fast_math", + "--threads=1", + "-Xptxas=-v", + "-diag-suppress=174", +] +cuda-minver = "12.6" +cxx-flags = [ + "-g", + "-O3", + "-fopenmp", + "-lgomp", + "-std=c++17", + "-DENABLE_BF16", ] +depends = ["torch"] include = ["."] src = [ - "sage_attention/fused/fused.cu", - "sage_attention/fused/fused.h" + "sage_attention/qattn/qk_int_sv_f16_cuda_sm80.cu", + "sage_attention/qattn/attn_cuda_sm80.h", + "sage_attention/qattn/attn_utils.cuh", +] + +[kernel._fused] +backend = "cuda" +cuda-capabilities = [ + "8.0", + "8.9", + "9.0a", ] -cxx-flags = ["-g", "-O3", "-fopenmp", "-lgomp", "-std=c++17", "-DENABLE_BF16"] cuda-flags = [ "-O3", "-std=c++17", @@ -145,6 +160,20 @@ cuda-flags = [ "--use_fast_math", "--threads=1", "-Xptxas=-v", - "-diag-suppress=174", # suppress the specific warning + "-diag-suppress=174", +] +cuda-minver = "12.6" +cxx-flags = [ + "-g", + "-O3", + "-fopenmp", + "-lgomp", + "-std=c++17", + "-DENABLE_BF16", +] +depends = ["torch"] +include = ["."] +src = [ + "sage_attention/fused/fused.cu", + "sage_attention/fused/fused.h", ] - diff --git a/build/torch28-cxx11-cu126-x86_64-linux/__init__.py b/build/torch28-cxx11-cu126-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..43b5e35a4154627e6457993372bd46b2cf78b89a --- /dev/null +++ b/build/torch28-cxx11-cu126-x86_64-linux/__init__.py @@ -0,0 +1,11 @@ +from .quant import per_block_int8, per_warp_int8, sub_mean, per_channel_fp8 +from .core import sageattn + + +__all__ = [ + "per_block_int8", + "per_warp_int8", + "sub_mean", + "per_channel_fp8", + "sageattn", +] \ No newline at end of file diff --git a/build/torch28-cxx11-cu126-x86_64-linux/_ops.py b/build/torch28-cxx11-cu126-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..e5a79964f03d3e86fffc66c151aa59b858911e88 --- /dev/null +++ b/build/torch28-cxx11-cu126-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _sage_attention_c08f355_dirty +ops = torch.ops._sage_attention_c08f355_dirty + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_sage_attention_c08f355_dirty::{op_name}" \ No newline at end of file diff --git a/build/torch28-cxx11-cu126-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so b/build/torch28-cxx11-cu126-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so new file mode 100755 index 0000000000000000000000000000000000000000..f5619c0bdc9f2040e482f7f2dd51dfe92d5ea22a --- /dev/null +++ b/build/torch28-cxx11-cu126-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5797f12cc80b2abf6b4bc12ede219be759ae9264452ebabcd8617be7111e4550 +size 26042464 diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/core.py b/build/torch28-cxx11-cu126-x86_64-linux/core.py similarity index 99% rename from build/torch28-cxx11-cu126-x86_64-linux/sage_attention/core.py rename to build/torch28-cxx11-cu126-x86_64-linux/core.py index 590e664e325d79068a56ee796a548f92ead07442..dadc6160869d0697d9752ab068593069d6aa4afa 100644 --- a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/core.py +++ b/build/torch28-cxx11-cu126-x86_64-linux/core.py @@ -116,7 +116,7 @@ def sageattn( - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` - All tensors must be on the same cuda device. """ - + print("########tensor_layout", tensor_layout) arch = get_cuda_arch_versions()[q.device.index] if arch == "sm80": return sageattn_qk_int8_pv_fp16_cuda( diff --git a/build/torch28-cxx11-cu126-x86_64-linux/metadata.json b/build/torch28-cxx11-cu126-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..76bafa5f33b6818aa6bb4cab04be811b87519b44 --- /dev/null +++ b/build/torch28-cxx11-cu126-x86_64-linux/metadata.json @@ -0,0 +1 @@ +{"python-depends":[]} \ No newline at end of file diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/quant.py b/build/torch28-cxx11-cu126-x86_64-linux/quant.py similarity index 100% rename from build/torch28-cxx11-cu126-x86_64-linux/sage_attention/quant.py rename to build/torch28-cxx11-cu126-x86_64-linux/quant.py diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/quant_per_thread.py b/build/torch28-cxx11-cu126-x86_64-linux/quant_per_thread.py similarity index 100% rename from build/torch28-cxx11-cu126-x86_64-linux/sage_attention/quant_per_thread.py rename to build/torch28-cxx11-cu126-x86_64-linux/quant_per_thread.py diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__init__.py b/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__init__.py index 43b5e35a4154627e6457993372bd46b2cf78b89a..03dbc1afe1cf156661a2b1b22003cd5f599a0309 100644 --- a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__init__.py +++ b/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__init__.py @@ -1,11 +1,26 @@ -from .quant import per_block_int8, per_warp_int8, sub_mean, per_channel_fp8 -from .core import sageattn +import ctypes +import sys +import importlib +from pathlib import Path +from types import ModuleType -__all__ = [ - "per_block_int8", - "per_warp_int8", - "sub_mean", - "per_channel_fp8", - "sageattn", -] \ No newline at end of file +def _import_from_path(file_path: Path) -> ModuleType: + # We cannot use the module name as-is, after adding it to `sys.modules`, + # it would also be used for other imports. So, we make a module name that + # depends on the path for it to be unique using the hex-encoded hash of + # the path. + path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value) + module_name = path_hash + spec = importlib.util.spec_from_file_location(module_name, file_path) + if spec is None: + raise ImportError(f"Cannot load spec for {module_name} from {file_path}") + module = importlib.util.module_from_spec(spec) + if module is None: + raise ImportError(f"Cannot load module {module_name} from spec") + sys.modules[module_name] = module + spec.loader.exec_module(module) # type: ignore + return module + + +globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py"))) diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/__init__.cpython-313.pyc b/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/__init__.cpython-313.pyc deleted file mode 100644 index 1265f608f89467b973713f14bc9ebc137ba1fe95..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/__init__.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/_ops.cpython-313.pyc b/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/_ops.cpython-313.pyc deleted file mode 100644 index c72bbe15717422f146f59a8c542d1fd40d477af2..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/_ops.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/core.cpython-313.pyc b/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/core.cpython-313.pyc deleted file mode 100644 index 2ec52ba13bf849b28c75a35aad0247fb80641c63..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/core.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/quant.cpython-313.pyc b/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/quant.cpython-313.pyc deleted file mode 100644 index d52ec4ede3e9be38fa11d10f593d65b9bbd9327e..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/quant.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/quant_per_thread.cpython-313.pyc b/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/quant_per_thread.cpython-313.pyc deleted file mode 100644 index 6f68872ec0cedce41a8abf3c3cfbd2d4e087e488..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/__pycache__/quant_per_thread.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/_ops.py b/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/_ops.py deleted file mode 100644 index 3f5bf8b095244a566572d3b7a9045ec0408db4e5..0000000000000000000000000000000000000000 --- a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/_ops.py +++ /dev/null @@ -1,9 +0,0 @@ -import torch -from . import _sage_attention_1369690_dirty -ops = torch.ops._sage_attention_1369690_dirty - -def add_op_namespace_prefix(op_name: str): - """ - Prefix op by namespace. - """ - return f"_sage_attention_1369690_dirty::{op_name}" \ No newline at end of file diff --git a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/_sage_attention_1369690_dirty.abi3.so b/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/_sage_attention_1369690_dirty.abi3.so deleted file mode 100755 index faac415c4ddf55aaa66929bbb468fe9edba9f1e1..0000000000000000000000000000000000000000 --- a/build/torch28-cxx11-cu126-x86_64-linux/sage_attention/_sage_attention_1369690_dirty.abi3.so +++ /dev/null @@ -1,3 +0,0 @@ -version https://git-lfs.github.com/spec/v1 -oid sha256:636461b53d3b3c4c1cd1940bc6ecb32728cb0f80bb347cf52591afd0ea121c8c -size 26037392 diff --git a/build/torch28-cxx11-cu128-x86_64-linux/__init__.py b/build/torch28-cxx11-cu128-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..43b5e35a4154627e6457993372bd46b2cf78b89a --- /dev/null +++ b/build/torch28-cxx11-cu128-x86_64-linux/__init__.py @@ -0,0 +1,11 @@ +from .quant import per_block_int8, per_warp_int8, sub_mean, per_channel_fp8 +from .core import sageattn + + +__all__ = [ + "per_block_int8", + "per_warp_int8", + "sub_mean", + "per_channel_fp8", + "sageattn", +] \ No newline at end of file diff --git a/build/torch28-cxx11-cu128-x86_64-linux/_ops.py b/build/torch28-cxx11-cu128-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..e5a79964f03d3e86fffc66c151aa59b858911e88 --- /dev/null +++ b/build/torch28-cxx11-cu128-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _sage_attention_c08f355_dirty +ops = torch.ops._sage_attention_c08f355_dirty + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_sage_attention_c08f355_dirty::{op_name}" \ No newline at end of file diff --git a/build/torch28-cxx11-cu128-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so b/build/torch28-cxx11-cu128-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so new file mode 100755 index 0000000000000000000000000000000000000000..9b9d0a6eabdf743ce6500445187d8c3caf323200 --- /dev/null +++ b/build/torch28-cxx11-cu128-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5f9b9a3becda02514dcb9ed79c5ea2a157cde2b6e63e21be55aa0bc52383892 +size 26554912 diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/core.py b/build/torch28-cxx11-cu128-x86_64-linux/core.py similarity index 99% rename from build/torch28-cxx11-cu129-x86_64-linux/sage_attention/core.py rename to build/torch28-cxx11-cu128-x86_64-linux/core.py index 590e664e325d79068a56ee796a548f92ead07442..dadc6160869d0697d9752ab068593069d6aa4afa 100644 --- a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/core.py +++ b/build/torch28-cxx11-cu128-x86_64-linux/core.py @@ -116,7 +116,7 @@ def sageattn( - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` - All tensors must be on the same cuda device. """ - + print("########tensor_layout", tensor_layout) arch = get_cuda_arch_versions()[q.device.index] if arch == "sm80": return sageattn_qk_int8_pv_fp16_cuda( diff --git a/build/torch28-cxx11-cu128-x86_64-linux/metadata.json b/build/torch28-cxx11-cu128-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..76bafa5f33b6818aa6bb4cab04be811b87519b44 --- /dev/null +++ b/build/torch28-cxx11-cu128-x86_64-linux/metadata.json @@ -0,0 +1 @@ +{"python-depends":[]} \ No newline at end of file diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/quant.py b/build/torch28-cxx11-cu128-x86_64-linux/quant.py similarity index 100% rename from build/torch28-cxx11-cu128-x86_64-linux/sage_attention/quant.py rename to build/torch28-cxx11-cu128-x86_64-linux/quant.py diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/quant_per_thread.py b/build/torch28-cxx11-cu128-x86_64-linux/quant_per_thread.py similarity index 100% rename from build/torch28-cxx11-cu128-x86_64-linux/sage_attention/quant_per_thread.py rename to build/torch28-cxx11-cu128-x86_64-linux/quant_per_thread.py diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__init__.py b/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__init__.py index 43b5e35a4154627e6457993372bd46b2cf78b89a..03dbc1afe1cf156661a2b1b22003cd5f599a0309 100644 --- a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__init__.py +++ b/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__init__.py @@ -1,11 +1,26 @@ -from .quant import per_block_int8, per_warp_int8, sub_mean, per_channel_fp8 -from .core import sageattn +import ctypes +import sys +import importlib +from pathlib import Path +from types import ModuleType -__all__ = [ - "per_block_int8", - "per_warp_int8", - "sub_mean", - "per_channel_fp8", - "sageattn", -] \ No newline at end of file +def _import_from_path(file_path: Path) -> ModuleType: + # We cannot use the module name as-is, after adding it to `sys.modules`, + # it would also be used for other imports. So, we make a module name that + # depends on the path for it to be unique using the hex-encoded hash of + # the path. + path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value) + module_name = path_hash + spec = importlib.util.spec_from_file_location(module_name, file_path) + if spec is None: + raise ImportError(f"Cannot load spec for {module_name} from {file_path}") + module = importlib.util.module_from_spec(spec) + if module is None: + raise ImportError(f"Cannot load module {module_name} from spec") + sys.modules[module_name] = module + spec.loader.exec_module(module) # type: ignore + return module + + +globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py"))) diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/__init__.cpython-313.pyc b/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/__init__.cpython-313.pyc deleted file mode 100644 index 456b0fbf7a0ba3ec6a2dc36e26287b1a6944c130..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/__init__.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/_ops.cpython-313.pyc b/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/_ops.cpython-313.pyc deleted file mode 100644 index 163ddb9c51de1f8e9df42d7a05dbb63fbe5313b0..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/_ops.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/core.cpython-313.pyc b/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/core.cpython-313.pyc deleted file mode 100644 index a3945ef7e2336f01d729258df5003ef84fa7449a..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/core.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/quant.cpython-313.pyc b/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/quant.cpython-313.pyc deleted file mode 100644 index a6b1e337187a955c72c026fa0c5e56ed15b65b99..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/quant.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/quant_per_thread.cpython-313.pyc b/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/quant_per_thread.cpython-313.pyc deleted file mode 100644 index e15aa4d4625174af8fc475ae5b5db3b6ed04f1be..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/__pycache__/quant_per_thread.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/_ops.py b/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/_ops.py deleted file mode 100644 index 3f5bf8b095244a566572d3b7a9045ec0408db4e5..0000000000000000000000000000000000000000 --- a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/_ops.py +++ /dev/null @@ -1,9 +0,0 @@ -import torch -from . import _sage_attention_1369690_dirty -ops = torch.ops._sage_attention_1369690_dirty - -def add_op_namespace_prefix(op_name: str): - """ - Prefix op by namespace. - """ - return f"_sage_attention_1369690_dirty::{op_name}" \ No newline at end of file diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/_sage_attention_1369690_dirty.abi3.so b/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/_sage_attention_1369690_dirty.abi3.so deleted file mode 100755 index 70be665802f3e6d80bab87150ebf7cbfc7141622..0000000000000000000000000000000000000000 --- a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/_sage_attention_1369690_dirty.abi3.so +++ /dev/null @@ -1,3 +0,0 @@ -version https://git-lfs.github.com/spec/v1 -oid sha256:2bfc9cab8f63665571d07c111a94edb3bec9a17aba7721a4c67be5392db0841d -size 26553920 diff --git a/build/torch28-cxx11-cu129-x86_64-linux/__init__.py b/build/torch28-cxx11-cu129-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..43b5e35a4154627e6457993372bd46b2cf78b89a --- /dev/null +++ b/build/torch28-cxx11-cu129-x86_64-linux/__init__.py @@ -0,0 +1,11 @@ +from .quant import per_block_int8, per_warp_int8, sub_mean, per_channel_fp8 +from .core import sageattn + + +__all__ = [ + "per_block_int8", + "per_warp_int8", + "sub_mean", + "per_channel_fp8", + "sageattn", +] \ No newline at end of file diff --git a/build/torch28-cxx11-cu129-x86_64-linux/_ops.py b/build/torch28-cxx11-cu129-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..e5a79964f03d3e86fffc66c151aa59b858911e88 --- /dev/null +++ b/build/torch28-cxx11-cu129-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _sage_attention_c08f355_dirty +ops = torch.ops._sage_attention_c08f355_dirty + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_sage_attention_c08f355_dirty::{op_name}" \ No newline at end of file diff --git a/build/torch28-cxx11-cu129-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so b/build/torch28-cxx11-cu129-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so new file mode 100755 index 0000000000000000000000000000000000000000..acd0b2284df0c175704c49b23643ef2d9254e882 --- /dev/null +++ b/build/torch28-cxx11-cu129-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e2eee18f99d9a6aa6d9179fbdd539c8653eba9a3692f35ad52a6fa07f99059f1 +size 26617232 diff --git a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/core.py b/build/torch28-cxx11-cu129-x86_64-linux/core.py similarity index 99% rename from build/torch28-cxx11-cu128-x86_64-linux/sage_attention/core.py rename to build/torch28-cxx11-cu129-x86_64-linux/core.py index 590e664e325d79068a56ee796a548f92ead07442..dadc6160869d0697d9752ab068593069d6aa4afa 100644 --- a/build/torch28-cxx11-cu128-x86_64-linux/sage_attention/core.py +++ b/build/torch28-cxx11-cu129-x86_64-linux/core.py @@ -116,7 +116,7 @@ def sageattn( - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` - All tensors must be on the same cuda device. """ - + print("########tensor_layout", tensor_layout) arch = get_cuda_arch_versions()[q.device.index] if arch == "sm80": return sageattn_qk_int8_pv_fp16_cuda( diff --git a/build/torch28-cxx11-cu129-x86_64-linux/metadata.json b/build/torch28-cxx11-cu129-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..76bafa5f33b6818aa6bb4cab04be811b87519b44 --- /dev/null +++ b/build/torch28-cxx11-cu129-x86_64-linux/metadata.json @@ -0,0 +1 @@ +{"python-depends":[]} \ No newline at end of file diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/quant.py b/build/torch28-cxx11-cu129-x86_64-linux/quant.py similarity index 100% rename from build/torch28-cxx11-cu129-x86_64-linux/sage_attention/quant.py rename to build/torch28-cxx11-cu129-x86_64-linux/quant.py diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/quant_per_thread.py b/build/torch28-cxx11-cu129-x86_64-linux/quant_per_thread.py similarity index 100% rename from build/torch28-cxx11-cu129-x86_64-linux/sage_attention/quant_per_thread.py rename to build/torch28-cxx11-cu129-x86_64-linux/quant_per_thread.py diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__init__.py b/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__init__.py index 43b5e35a4154627e6457993372bd46b2cf78b89a..03dbc1afe1cf156661a2b1b22003cd5f599a0309 100644 --- a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__init__.py +++ b/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__init__.py @@ -1,11 +1,26 @@ -from .quant import per_block_int8, per_warp_int8, sub_mean, per_channel_fp8 -from .core import sageattn +import ctypes +import sys +import importlib +from pathlib import Path +from types import ModuleType -__all__ = [ - "per_block_int8", - "per_warp_int8", - "sub_mean", - "per_channel_fp8", - "sageattn", -] \ No newline at end of file +def _import_from_path(file_path: Path) -> ModuleType: + # We cannot use the module name as-is, after adding it to `sys.modules`, + # it would also be used for other imports. So, we make a module name that + # depends on the path for it to be unique using the hex-encoded hash of + # the path. + path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value) + module_name = path_hash + spec = importlib.util.spec_from_file_location(module_name, file_path) + if spec is None: + raise ImportError(f"Cannot load spec for {module_name} from {file_path}") + module = importlib.util.module_from_spec(spec) + if module is None: + raise ImportError(f"Cannot load module {module_name} from spec") + sys.modules[module_name] = module + spec.loader.exec_module(module) # type: ignore + return module + + +globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py"))) diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/__init__.cpython-313.pyc b/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/__init__.cpython-313.pyc deleted file mode 100644 index 77ee0c5cb4d71e324a08437085329d613bddb942..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/__init__.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/_ops.cpython-313.pyc b/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/_ops.cpython-313.pyc deleted file mode 100644 index 3304fc1253a1ff2b04ec288f556cfff1ecc2ca92..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/_ops.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/core.cpython-313.pyc b/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/core.cpython-313.pyc deleted file mode 100644 index 1feb3f559f17390d27dc3266976cfbefc3c83c1e..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/core.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/quant.cpython-313.pyc b/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/quant.cpython-313.pyc deleted file mode 100644 index e7e00ede7b0bf73713e2545e8c3b136e3b521290..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/quant.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/quant_per_thread.cpython-313.pyc b/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/quant_per_thread.cpython-313.pyc deleted file mode 100644 index ffaef41899636fbeae8a2725f47d9b70074f7ed3..0000000000000000000000000000000000000000 Binary files a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/__pycache__/quant_per_thread.cpython-313.pyc and /dev/null differ diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/_ops.py b/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/_ops.py deleted file mode 100644 index 3f5bf8b095244a566572d3b7a9045ec0408db4e5..0000000000000000000000000000000000000000 --- a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/_ops.py +++ /dev/null @@ -1,9 +0,0 @@ -import torch -from . import _sage_attention_1369690_dirty -ops = torch.ops._sage_attention_1369690_dirty - -def add_op_namespace_prefix(op_name: str): - """ - Prefix op by namespace. - """ - return f"_sage_attention_1369690_dirty::{op_name}" \ No newline at end of file diff --git a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/_sage_attention_1369690_dirty.abi3.so b/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/_sage_attention_1369690_dirty.abi3.so deleted file mode 100755 index 3a4768d30801b4411c86872d8fe296d0a27161ff..0000000000000000000000000000000000000000 --- a/build/torch28-cxx11-cu129-x86_64-linux/sage_attention/_sage_attention_1369690_dirty.abi3.so +++ /dev/null @@ -1,3 +0,0 @@ -version https://git-lfs.github.com/spec/v1 -oid sha256:fef61629b7537ad41d31cf5715a11a38ce3f7cc97b0d5bf26356492b36ad5c29 -size 26608048 diff --git a/build/torch29-cxx11-cu126-x86_64-linux/__init__.py b/build/torch29-cxx11-cu126-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..43b5e35a4154627e6457993372bd46b2cf78b89a --- /dev/null +++ b/build/torch29-cxx11-cu126-x86_64-linux/__init__.py @@ -0,0 +1,11 @@ +from .quant import per_block_int8, per_warp_int8, sub_mean, per_channel_fp8 +from .core import sageattn + + +__all__ = [ + "per_block_int8", + "per_warp_int8", + "sub_mean", + "per_channel_fp8", + "sageattn", +] \ No newline at end of file diff --git a/build/torch29-cxx11-cu126-x86_64-linux/_ops.py b/build/torch29-cxx11-cu126-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..e5a79964f03d3e86fffc66c151aa59b858911e88 --- /dev/null +++ b/build/torch29-cxx11-cu126-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _sage_attention_c08f355_dirty +ops = torch.ops._sage_attention_c08f355_dirty + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_sage_attention_c08f355_dirty::{op_name}" \ No newline at end of file diff --git a/build/torch29-cxx11-cu126-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so b/build/torch29-cxx11-cu126-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so new file mode 100755 index 0000000000000000000000000000000000000000..231470c441a65b79d776e5f7d6d38ea3c316693b --- /dev/null +++ b/build/torch29-cxx11-cu126-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c8eeda7b9569b9b3d961438b0dee1920deb2324888ae2e15e27251f989ad4e8 +size 26046912 diff --git a/build/torch29-cxx11-cu126-x86_64-linux/core.py b/build/torch29-cxx11-cu126-x86_64-linux/core.py new file mode 100644 index 0000000000000000000000000000000000000000..dadc6160869d0697d9752ab068593069d6aa4afa --- /dev/null +++ b/build/torch29-cxx11-cu126-x86_64-linux/core.py @@ -0,0 +1,969 @@ +""" +Copyright (c) 2024 by SageAttention team. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +""" + +import torch +import torch.nn.functional as F + +from ._ops import ops + + +from .quant import per_warp_int8 as per_warp_int8_cuda +from .quant import sub_mean +from .quant import per_channel_fp8 +from .quant_per_thread import per_thread_int8 as per_thread_int8_triton + +from typing import Any, List, Literal, Optional, Tuple, Union +import warnings + + +import subprocess +import re + + +def get_cuda_version(): + try: + output = subprocess.check_output(["nvcc", "--version"]).decode() + match = re.search(r"release (\d+)\.(\d+)", output) + if match: + major, minor = int(match.group(1)), int(match.group(2)) + return major, minor + except Exception as e: + print("Failed to get CUDA version:", e) + return None, None + + +def get_cuda_arch_versions(): + cuda_archs = [] + for i in range(torch.cuda.device_count()): + major, minor = torch.cuda.get_device_capability(i) + cuda_archs.append(f"sm{major}{minor}") + return cuda_archs + + +def sageattn( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + sm_scale: Optional[float] = None, + return_lse: bool = False, + **kwargs: Any, +): + """ + Automatically selects the appropriate implementation of the SageAttention kernel based on the GPU compute capability. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + """ + print("########tensor_layout", tensor_layout) + arch = get_cuda_arch_versions()[q.device.index] + if arch == "sm80": + return sageattn_qk_int8_pv_fp16_cuda( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32", + ) + elif arch == "sm89": + return sageattn_qk_int8_pv_fp8_cuda( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32+fp16", + ) + elif arch == "sm90": + return sageattn_qk_int8_pv_fp8_cuda_sm90( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32+fp32", + ) + elif arch == "sm120": + return sageattn_qk_int8_pv_fp8_cuda( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + qk_quant_gran="per_warp", + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32+fp16", + ) # sm120 has accurate fp32 accumulator for fp8 mma and triton kernel is currently not usable on sm120. + else: + raise ValueError(f"Unsupported CUDA architecture: {arch}") + + +@torch.compiler.disable +def sageattn_qk_int8_pv_fp16_cuda( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + qk_quant_gran: str = "per_thread", + sm_scale: Optional[float] = None, + pv_accum_dtype: str = "fp32", + smooth_k: bool = True, + smooth_v: bool = False, + return_lse: bool = False, + **kwargs: Any, +) -> torch.Tensor: + """ + SageAttention with INT8 quantization for Q and K, FP16 PV with FP16/FP32 accumulation, implemented using CUDA. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + qk_quant_gran : str + The granularity of quantization for Q and K, either "per_warp" or "per_thread". + Default: "per_thread". + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + pv_accum_dtype : str + The dtype of the accumulation of the product of the value tensor and the attention weights, either "fp16", "fp16+fp32" or "fp32". + - "fp16": PV accumulation is done in fully in FP16. This is the fastest option but may lead to numerical instability. `smooth_v` option will increase the accuracy in cases when the value tensor has a large bias (like in CogVideoX-2b). + - "fp32": PV accumulation is done in FP32. This is the most accurate option but may be slower than "fp16" due to CUDA core overhead. + - "fp16+fp32": PV accumulation is done in FP16, but added to a FP32 buffer every few iterations. This offers a balance between speed and accuracy. + Default: "fp32". + + smooth_k : bool + Whether to smooth the key tensor by subtracting the mean along the sequence dimension. + Default: True. + + smooth_v : bool + Whether to smooth the value tensor by subtracting the mean along the sequence dimension. + smooth_v will be ignored if pv_accum_dtype is "fp32" or "fp16+fp32". + Default: False. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + - `smooth_k` will introduce slight overhead but will improve the accuracy under most circumstances. + """ + + dtype = q.dtype + assert q.is_cuda, "Input tensors must be on cuda." + assert dtype in [torch.float16, torch.bfloat16], ( + "Input tensors must be in dtype of torch.float16 or torch.bfloat16" + ) + assert qk_quant_gran in ["per_warp", "per_thread"], ( + "qk_quant_gran must be either 'per_warp' or 'per_thread'." + ) + assert q.device == k.device == v.device, "All tensors must be on the same device." + assert q.dtype == k.dtype == v.dtype, "All tensors must have the same dtype." + + # FIXME(DefTruth): make sage attention work compatible with distributed + # env, for example, xDiT which launch by torchrun. Without this workaround, + # sage attention will run into illegal memory access error after first + # inference step in distributed env for multi gpus inference. This small + # workaround also make sage attention work compatible with torch.compile + # through non-fullgraph compile mode. + torch.cuda.set_device(v.device) + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + _is_caual = 1 if is_causal else 0 + _qk_quant_gran = 3 if qk_quant_gran == "per_thread" else 2 + _return_lse = 1 if return_lse else 0 + + head_dim_og = q.size(-1) + + if head_dim_og < 64: + q = torch.nn.functional.pad(q, (0, 64 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 64 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 64 - head_dim_og)) + elif head_dim_og > 64 and head_dim_og < 128: + q = torch.nn.functional.pad(q, (0, 128 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 128 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 128 - head_dim_og)) + elif head_dim_og > 128: + raise ValueError(f"Unsupported head_dim: {head_dim_og}") + + # assert last dim is contiguous + assert q.stride(-1) == 1 and k.stride(-1) == 1 and v.stride(-1) == 1, ( + "Last dim of qkv must be contiguous." + ) + + if sm_scale is None: + sm_scale = head_dim_og**-0.5 + + seq_dim = 1 if _tensor_layout == 0 else 2 + nh_dim = 2 if _tensor_layout == 0 else 1 + + if smooth_k: + km = k.mean(dim=seq_dim, keepdim=True) + nqheads = q.size(2) + nkheads = k.size(2) + q_per_kv_heads = nqheads // nkheads + if q_per_kv_heads > 1: + # nheads_k => nheads_q + km_broadcast = torch.repeat_interleave(km, q_per_kv_heads, dim=nh_dim) + else: + km_broadcast = km + if return_lse: + if tensor_layout == "NHD": + lse_correction = ( + torch.matmul( + q.transpose(1, 2), km_broadcast.transpose(1, 2).transpose(2, 3) + ) + .squeeze(-1) + .to(torch.float32) + ) + else: + lse_correction = ( + torch.matmul(q, km_broadcast.transpose(2, 3)) + .squeeze(-1) + .to(torch.float32) + ) + else: + km = None + + if qk_quant_gran == "per_warp": + q_int8, q_scale, k_int8, k_scale = per_warp_int8_cuda( + q, + k, + km, + tensor_layout=tensor_layout, + BLKQ=128, + WARPQ=(16 if (q.size(-1) == 128 and pv_accum_dtype == "fp16+fp32") else 32), + BLKK=64, + ) + elif qk_quant_gran == "per_thread": + q_int8, q_scale, k_int8, k_scale = per_thread_int8_triton( + q, + k, + km, + tensor_layout=tensor_layout, + BLKQ=128, + WARPQ=(16 if (q.size(-1) == 128 and pv_accum_dtype == "fp16+fp32") else 32), + BLKK=64, + WARPK=64, + ) + + o = torch.empty(q.size(), dtype=dtype, device=q.device) + + if pv_accum_dtype in ["fp32", "fp16+fp32"] and smooth_v: + warnings.warn(f"pv_accum_dtype is {pv_accum_dtype}, smooth_v will be ignored.") + smooth_v = False + + if pv_accum_dtype == "fp32": + v = v.to(torch.float16) + lse = ops.qk_int8_sv_f16_accum_f32_attn( + q_int8, + k_int8, + v, + o, + q_scale, + k_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + elif pv_accum_dtype == "fp16": + if smooth_v: + smoothed_v, vm = sub_mean(v, tensor_layout=tensor_layout) + lse = ops.qk_int8_sv_f16_accum_f16_fuse_v_mean_attn( + q_int8, + k_int8, + smoothed_v, + o, + q_scale, + k_scale, + vm, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + else: + v = v.to(torch.float16) + lse = ops.qk_int8_sv_f16_accum_f16_attn( + q_int8, + k_int8, + v, + o, + q_scale, + k_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + elif pv_accum_dtype == "fp16+fp32": + v = v.to(torch.float16) + lse = ops.qk_int8_sv_f16_accum_f16_attn_inst_buf( + q_int8, + k_int8, + v, + o, + q_scale, + k_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + else: + raise ValueError(f"Unsupported pv_accum_dtype: {pv_accum_dtype}") + + o = o[..., :head_dim_og] + + if return_lse: + return ( + o, + lse / 1.44269504 + lse_correction * sm_scale + if smooth_k + else lse / 1.44269504, + ) + else: + return o + + +@torch.compiler.disable +def sageattn_qk_int8_pv_fp8_cuda( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + qk_quant_gran: str = "per_thread", + sm_scale: Optional[float] = None, + pv_accum_dtype: str = "fp32+fp16", + smooth_k: bool = True, + smooth_v: bool = False, + return_lse: bool = False, + **kwargs: Any, +) -> torch.Tensor: + """ + SageAttention with INT8 quantization for Q and K, FP8 PV with FP32 accumulation, implemented using CUDA. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + qk_quant_gran : str + The granularity of quantization for Q and K, either "per_warp" or "per_thread". + Default: "per_thread". + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + pv_accum_dtype : str + The dtype of the accumulation of the product of the value tensor and the attention weights, either "fp32" or "fp32+fp32". + - "fp32": PV accumulation is done in fully in FP32. However, due to the hardware issue, there are only 22 valid bits in the FP32 accumulator. + - "fp32+fp32": PV accumulation is done in FP32 (actually FP22), but added to a FP32 buffer every few iterations. This offers a balance between speed and accuracy. + Default: "fp32+fp32". + + smooth_k : bool + Whether to smooth the key tensor by subtracting the mean along the sequence dimension. + Default: True. + + smooth_v : bool + Whether to smooth the value tensor by subtracting the mean along the sequence dimension. + smooth_v will be ignored if pv_accum_dtype is "fp32+fp32". + Default: False. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + - `smooth_k` will introduce slight overhead but will improve the accuracy under most circumstances. + """ + + dtype = q.dtype + assert q.is_cuda, "Input tensors must be on cuda." + assert dtype in [torch.float16, torch.bfloat16], ( + "Input tensors must be in dtype of torch.float16 or torch.bfloat16" + ) + assert qk_quant_gran in ["per_warp", "per_thread"], ( + "qk_quant_gran must be either 'per_warp' or 'per_thread'." + ) + assert q.device == k.device == v.device, "All tensors must be on the same device." + assert q.dtype == k.dtype == v.dtype, "All tensors must have the same dtype." + + # cuda_major_version, cuda_minor_version = get_cuda_version() + # if(cuda_major_version, cuda_minor_version) < (12, 8) and pv_accum_dtype == 'fp32+fp16': + # warnings.warn("cuda version < 12.8, change pv_accum_dtype to 'fp32+fp32'") + # pv_accum_dtype = 'fp32+fp32' + + # FIXME(DefTruth): make sage attention work compatible with distributed + # env, for example, xDiT which launch by torchrun. Without this workaround, + # sage attention will run into illegal memory access error after first + # inference step in distributed env for multi gpus inference. This small + # workaround also make sage attention work compatible with torch.compile + # through non-fullgraph compile mode. + torch.cuda.set_device(v.device) + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + _is_caual = 1 if is_causal else 0 + _qk_quant_gran = 3 if qk_quant_gran == "per_thread" else 2 + _return_lse = 1 if return_lse else 0 + + head_dim_og = q.size(-1) + + if head_dim_og < 64: + q = torch.nn.functional.pad(q, (0, 64 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 64 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 64 - head_dim_og)) + elif head_dim_og > 64 and head_dim_og < 128: + q = torch.nn.functional.pad(q, (0, 128 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 128 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 128 - head_dim_og)) + elif head_dim_og > 128: + raise ValueError(f"Unsupported head_dim: {head_dim_og}") + + # assert last dim is contiguous + assert q.stride(-1) == 1 and k.stride(-1) == 1 and v.stride(-1) == 1, ( + "Last dim of qkv must be contiguous." + ) + + if sm_scale is None: + sm_scale = head_dim_og**-0.5 + + seq_dim = 1 if _tensor_layout == 0 else 2 + nh_dim = 2 if _tensor_layout == 0 else 1 + + if smooth_k: + km = k.mean(dim=seq_dim, keepdim=True) + nqheads = q.size(2) + nkheads = k.size(2) + q_per_kv_heads = nqheads // nkheads + if q_per_kv_heads > 1: + # nheads_k => nheads_q + km_broadcast = torch.repeat_interleave(km, q_per_kv_heads, dim=nh_dim) + else: + km_broadcast = km + if return_lse: + if tensor_layout == "NHD": + lse_correction = ( + torch.matmul( + q.transpose(1, 2), km_broadcast.transpose(1, 2).transpose(2, 3) + ) + .squeeze(-1) + .to(torch.float32) + ) + else: + lse_correction = ( + torch.matmul(q, km_broadcast.transpose(2, 3)) + .squeeze(-1) + .to(torch.float32) + ) + else: + km = None + + if qk_quant_gran == "per_warp": + q_int8, q_scale, k_int8, k_scale = per_warp_int8_cuda( + q, k, km, tensor_layout=tensor_layout, BLKQ=128, WARPQ=32, BLKK=64 + ) + elif qk_quant_gran == "per_thread": + q_int8, q_scale, k_int8, k_scale = per_thread_int8_triton( + q, k, km, tensor_layout=tensor_layout, BLKQ=128, WARPQ=32, BLKK=64, WARPK=64 + ) + + o = torch.empty(q.size(), dtype=dtype, device=q.device) + + if pv_accum_dtype == "fp32+fp32" and smooth_v: + warnings.warn("pv_accum_dtype is 'fp32+fp32', smooth_v will be ignored.") + smooth_v = False + + if pv_accum_dtype == "fp32+fp16" and smooth_v: + warnings.warn("pv_accum_dtype is 'fp32+fp16', smooth_v will be ignored.") + smooth_v = False + + quant_v_scale_max = 448.0 + if pv_accum_dtype == "fp32+fp16": + quant_v_scale_max = 2.25 + + v_fp8, v_scale, vm = per_channel_fp8( + v, tensor_layout=tensor_layout, scale_max=quant_v_scale_max, smooth_v=smooth_v + ) + print("before kernel call") + if pv_accum_dtype == "fp32": + if smooth_v: + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_fuse_v_mean_attn( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + vm, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + else: + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + elif pv_accum_dtype == "fp32+fp32": + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + elif pv_accum_dtype == "fp32+fp16": + lse = ops.qk_int8_sv_f8_accum_f16_fuse_v_scale_attn_inst_buf( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + o = o[..., :head_dim_og] + print("after kernel call") + if return_lse: + return ( + o, + lse / 1.44269504 + lse_correction * sm_scale + if smooth_k + else lse / 1.44269504, + ) + else: + return o + + +@torch.compiler.disable +def sageattn_qk_int8_pv_fp8_cuda_sm90( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + qk_quant_gran: str = "per_thread", + sm_scale: Optional[float] = None, + pv_accum_dtype: str = "fp32+fp32", + smooth_k: bool = True, + return_lse: bool = False, + **kwargs: Any, +) -> torch.Tensor: + """ + SageAttention with INT8 quantization for Q and K, FP8 PV with FP32 accumulation, implemented using CUDA. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + qk_quant_gran : str + The granularity of quantization for Q and K, either "per_warp" or "per_thread". + Default: "per_thread". + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + pv_accum_dtype : str + The dtype of the accumulation of the product of the value tensor and the attention weights, either "fp32" or "fp32+fp32". + - "fp32": PV accumulation is done in fully in FP32. However, due to the hardware issue, there are only 22 valid bits in the FP32 accumulator. + - "fp32+fp32": PV accumulation is done in FP32 (actually FP22), but added to a FP32 buffer every few iterations. This offers a balance between speed and accuracy. + Default: "fp32+fp32". + + smooth_k : bool + Whether to smooth the key tensor by subtracting the mean along the sequence dimension. + Default: True. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + - `smooth_k` will introduce slight overhead but will improve the accuracy under most circumstances. + """ + + dtype = q.dtype + assert q.is_cuda, "Input tensors must be on cuda." + assert dtype in [torch.float16, torch.bfloat16], ( + "Input tensors must be in dtype of torch.float16 or torch.bfloat16" + ) + assert qk_quant_gran in ["per_warp", "per_thread"], ( + "qk_quant_gran must be either 'per_warp' or 'per_thread'." + ) + assert q.device == k.device == v.device, "All tensors must be on the same device." + assert q.dtype == k.dtype == v.dtype, "All tensors must have the same dtype." + + torch.cuda.set_device(v.device) + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + _is_caual = 1 if is_causal else 0 + _qk_quant_gran = 3 if qk_quant_gran == "per_thread" else 2 + _return_lse = 1 if return_lse else 0 + + head_dim_og = q.size(-1) + + if head_dim_og < 64: + q = torch.nn.functional.pad(q, (0, 64 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 64 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 64 - head_dim_og)) + elif head_dim_og > 64 and head_dim_og < 128: + q = torch.nn.functional.pad(q, (0, 128 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 128 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 128 - head_dim_og)) + elif head_dim_og > 128: + raise ValueError(f"Unsupported head_dim: {head_dim_og}") + + # assert last dim is contiguous + assert q.stride(-1) == 1 and k.stride(-1) == 1 and v.stride(-1) == 1, ( + "Last dim of qkv must be contiguous." + ) + + if sm_scale is None: + sm_scale = head_dim_og**-0.5 + + seq_dim = 1 if _tensor_layout == 0 else 2 + nh_dim = 2 if _tensor_layout == 0 else 1 + + if smooth_k: + km = k.mean(dim=seq_dim, keepdim=True) + nqheads = q.size(2) + nkheads = k.size(2) + q_per_kv_heads = nqheads // nkheads + if q_per_kv_heads > 1: + # nheads_k => nheads_q + km_broadcast = torch.repeat_interleave(km, q_per_kv_heads, dim=nh_dim) + else: + km_broadcast = km + if return_lse: + if tensor_layout == "NHD": + lse_correction = ( + torch.matmul( + q.transpose(1, 2), km_broadcast.transpose(1, 2).transpose(2, 3) + ) + .squeeze(-1) + .to(torch.float32) + ) + else: + lse_correction = ( + torch.matmul(q, km_broadcast.transpose(2, 3)) + .squeeze(-1) + .to(torch.float32) + ) + else: + km = None + + if qk_quant_gran == "per_warp": + q_int8, q_scale, k_int8, k_scale = per_warp_int8_cuda( + q, k, km, tensor_layout=tensor_layout, BLKQ=64, WARPQ=16, BLKK=128 + ) + elif qk_quant_gran == "per_thread": + q_int8, q_scale, k_int8, k_scale = per_thread_int8_triton( + q, + k, + km, + tensor_layout=tensor_layout, + BLKQ=64, + WARPQ=16, + BLKK=128, + WARPK=128, + ) + + o = torch.empty(q.size(), dtype=dtype, device=q.device) + + # pad v to multiple of 128 + # TODO: modify per_channel_fp8 kernel to handle this + kv_len = k.size(seq_dim) + v_pad_len = 128 - (kv_len % 128) if kv_len % 128 != 0 else 0 + if v_pad_len > 0: + if tensor_layout == "HND": + v = torch.cat( + [ + v, + torch.zeros( + v.size(0), + v.size(1), + v_pad_len, + v.size(3), + dtype=v.dtype, + device=v.device, + ), + ], + dim=2, + ) + else: + v = torch.cat( + [ + v, + torch.zeros( + v.size(0), + v_pad_len, + v.size(2), + v.size(3), + dtype=v.dtype, + device=v.device, + ), + ], + dim=1, + ) + + v_fp8, v_scale, _ = per_channel_fp8(v, tensor_layout=tensor_layout, smooth_v=False) + + if pv_accum_dtype == "fp32": + raise NotImplementedError("Please use pv_accum_dtype='fp32+fp32' for sm90.") + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + elif pv_accum_dtype == "fp32+fp32": + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf_sm90( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + + o = o[..., :head_dim_og] + + if return_lse: + return ( + o, + lse / 1.44269504 + lse_correction * sm_scale + if smooth_k + else lse / 1.44269504, + ) + else: + return o diff --git a/build/torch29-cxx11-cu126-x86_64-linux/metadata.json b/build/torch29-cxx11-cu126-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..76bafa5f33b6818aa6bb4cab04be811b87519b44 --- /dev/null +++ b/build/torch29-cxx11-cu126-x86_64-linux/metadata.json @@ -0,0 +1 @@ +{"python-depends":[]} \ No newline at end of file diff --git a/build/torch29-cxx11-cu126-x86_64-linux/quant.py b/build/torch29-cxx11-cu126-x86_64-linux/quant.py new file mode 100644 index 0000000000000000000000000000000000000000..2e7a32c6a1e502bee12d0e0564ff2b90f6b00462 --- /dev/null +++ b/build/torch29-cxx11-cu126-x86_64-linux/quant.py @@ -0,0 +1,326 @@ +""" +Copyright (c) 2024 by SageAttention team. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +""" + +import torch +from typing import Optional + +from ._ops import ops + + +def per_block_int8( + q: torch.Tensor, + k: torch.Tensor, + km: Optional[torch.Tensor] = None, + BLKQ: int = 128, + BLKK: int = 64, + sm_scale: Optional[float] = None, + tensor_layout: str = "HND", +): + """ + Quantize the query tensor `q` and the key tensor `k` with per block quantization. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + km : Optional[torch.Tensor] + The mean tensor of `k` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]``. + Should be of the same dtype as `k` if provided. Default is None. + + sm_scale : Optional[float] + The scale factor for the softmax operation. Default is ``head_dim**-0.5``. + It will be multiplied by ``1.44269504`` to work together with the triton attention kernel. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor] + A tuple containing: + - The quantized query tensor. Shape: Same as `q` but with `int8` dtype. + - The scale tensor of the query tensor. Shape: ``[batch_size, num_qo_heads, (qo_len + BLKQ - 1) // BLKQ]`` with `float32` dtype. + - The quantized key tensor. Shape: Same as `k` but with `int8` dtype. + - The scale tensor of the key tensor. Shape: ``[batch_size, num_kv_heads, (kv_len + BLKK - 1) // BLKK]`` with `float32` dtype. + + Note + ---- + - The tensors `q` and `k` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + """ + + q_int8 = torch.empty(q.shape, dtype=torch.int8, device=q.device) + k_int8 = torch.empty(k.shape, dtype=torch.int8, device=k.device) + + if tensor_layout == "HND": + b, h_qo, qo_len, head_dim = q.shape + _, h_kv, kv_len, _ = k.shape + + elif tensor_layout == "NHD": + b, qo_len, h_qo, head_dim = q.shape + _, kv_len, h_kv, _ = k.shape + + else: + raise ValueError(f"Unknown tensor layout: {tensor_layout}") + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + + q_scale = torch.empty( + (b, h_qo, (qo_len + BLKQ - 1) // BLKQ), device=q.device, dtype=torch.float32 + ) + k_scale = torch.empty( + (b, h_kv, (kv_len + BLKK - 1) // BLKK), device=q.device, dtype=torch.float32 + ) + + if sm_scale is None: + sm_scale = head_dim**-0.5 + + sm_scale *= 1.44269504 + + ops.quant_per_block_int8_cuda(q, q_int8, q_scale, sm_scale, BLKQ, _tensor_layout) + if km is not None: + km = km.squeeze(1) if _tensor_layout == 0 else km.squeeze(2) + ops.quant_per_block_int8_fuse_sub_mean_cuda( + k, km, k_int8, k_scale, BLKK, _tensor_layout + ) + else: + # The bound CUDA op expects an sm_scale argument; use 1.0 for K to avoid scaling + ops.quant_per_block_int8_cuda(k, k_int8, k_scale, 1.0, BLKK, _tensor_layout) + + return q_int8, q_scale, k_int8, k_scale + + +def per_warp_int8( + q: torch.Tensor, + k: torch.Tensor, + km: Optional[torch.Tensor] = None, + BLKQ: int = 128, + WARPQ: int = 32, + BLKK: int = 64, + tensor_layout: str = "HND", +): + """ + Quantize the query tensor `q` with per warp quantization and the key tensor `k` with per block quantization. + Warp size of quantizing `q` is 16 or 32, with a block size of 64 or 128. + Block size of quantizing `k` is 64 or 128. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + km : Optional[torch.Tensor] + The mean tensor of `k` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]``. + Should be of the same dtype as `k` if provided. Default is None. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor] + A tuple containing: + - The quantized query tensor. Shape: Same as `q` but with `int8` dtype. + - The scale tensor of the query tensor. Shape: ``[batch_size, num_qo_heads, (qo_len + BLKQ - 1) // BLKQ * (BLKQ // WARPQ)]`` with `float32` dtype. + - The quantized key tensor. Shape: Same as `k` but with `int8` dtype. + - The scale tensor of the key tensor. Shape: ``[batch_size, num_kv_heads, (kv_len + BLKK - 1) // BLKK]`` with `float32` dtype. + + Note + ---- + - The tensors `q` and `k` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + """ + + q_int8 = torch.empty(q.shape, dtype=torch.int8, device=q.device) + k_int8 = torch.empty(k.shape, dtype=torch.int8, device=k.device) + + if tensor_layout == "HND": + b, h_qo, qo_len, head_dim = q.shape + _, h_kv, kv_len, _ = k.shape + + elif tensor_layout == "NHD": + b, qo_len, h_qo, head_dim = q.shape + _, kv_len, h_kv, _ = k.shape + + else: + raise ValueError(f"Unknown tensor layout: {tensor_layout}") + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + + q_scale = torch.empty( + (b, h_qo, ((qo_len + BLKQ - 1) // BLKQ) * (BLKQ // WARPQ)), + device=q.device, + dtype=torch.float32, + ) + k_scale = torch.empty( + (b, h_kv, (kv_len + BLKK - 1) // BLKK), device=q.device, dtype=torch.float32 + ) + + ops.quant_per_warp_int8_cuda(q, q_int8, q_scale, BLKQ, WARPQ, _tensor_layout) + + if km is not None: + km = km.squeeze(1) if _tensor_layout == 0 else km.squeeze(2) + ops.quant_per_block_int8_fuse_sub_mean_cuda( + k, km, k_int8, k_scale, BLKK, _tensor_layout + ) + else: + # The bound CUDA op expects an sm_scale argument; use 1.0 for K to avoid scaling + ops.quant_per_block_int8_cuda(k, k_int8, k_scale, 1.0, BLKK, _tensor_layout) + + return q_int8, q_scale, k_int8, k_scale + + +def sub_mean(v: torch.Tensor, tensor_layout: str = "HND"): + """ + Calculate the mean of the tensor `v` along the sequence length dimension and subtract it from `v`. Result is stored as fp16. + + Parameters + ---------- + v : torch.Tensor + The input tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor] + A tuple containing: + - The tensor `v_smoothed` with the mean subtracted and stored as fp16. Shape: Same as `v` with `float16` dtype. + - The mean tensor of `v` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]`` with dtype same as `v`. + + Note + ---- + - The tensors `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - The returned tensor `v_smoothed` will have dtype ``torch.float16`` regardless of the input dtype. + - The returned mean tensor will have the same dtype as the input tensor. + """ + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + vm = v.mean(dim=1 if _tensor_layout == 0 else 2) + + v_smoothed = torch.empty(v.shape, dtype=torch.float16, device=v.device) + + # subtract mean and store the result as fp16 + ops.sub_mean_cuda(v, vm, v_smoothed, _tensor_layout) + + return v_smoothed, vm + + +def per_channel_fp8( + v: torch.Tensor, + tensor_layout: str = "HND", + scale_max: float = 448.0, + smooth_v: bool = True, +): + """ + Transpose, pad and permute the tensor `v` and quantize it to fp8 with per channel quantization. + `v` is first transposed along the head dimension and the sequence length dimension, then padded to a multiple of 64. + After that, the tensor is permuted along the sequence length dimension by ``[0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15]``. + The quantization is done per channel, with the scale value and smooth factor calculated per channel. + + Parameters + ---------- + v : torch.Tensor + The input tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + scale_max : float + The maximum scale value for the quantization. Default is 448.0 (upper bound of E4M3 data format). + + smooth_v : bool + Whether to smooth the quantized tensor. Default is True. + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor, Optional[torch.Tensor]] + A tuple containing: + - The quantized tensor `v_fp8`. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, head_dim, (kv_len + 63) // 64 * 64]``, with `float8_e4m3fn` dtype. + - If `tensor_layout` is "NHD": ``[batch_size, head_dim, num_kv_heads, (kv_len + 63) // 64 * 64]``, with `float8_e4m3fn` dtype. + - The scale tensor of `v`. Shape: ``[batch_size, num_kv_heads, head_dim]`` with `float32` dtype. + - The mean tensor of `v` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]`` with `float32` dtype. + + Note + ---- + - The tensors `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - The returned mean tensor will be None if `smooth_v` is False. Otherwise it will have dtype ``torch.float32``. + """ + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + + if tensor_layout == "HND": + b, h_kv, kv_len, head_dim = v.shape + padded_len = (kv_len + 63) // 64 * 64 + v_transposed_permutted = torch.empty( + (b, h_kv, head_dim, padded_len), dtype=v.dtype, device=v.device + ) + + elif tensor_layout == "NHD": + b, kv_len, h_kv, head_dim = v.shape + padded_len = (kv_len + 63) // 64 * 64 + v_transposed_permutted = torch.empty( + (b, head_dim, h_kv, padded_len), dtype=v.dtype, device=v.device + ) + + ops.transpose_pad_permute_cuda(v, v_transposed_permutted, _tensor_layout) + + v_fp8 = torch.empty( + v_transposed_permutted.shape, dtype=torch.float8_e4m3fn, device=v.device + ) + + v_scale = torch.empty((b, h_kv, head_dim), dtype=torch.float32, device=v.device) + vm = torch.empty((b, h_kv, head_dim), dtype=torch.float32, device=v.device) + + if smooth_v: + ops.mean_scale_fuse_quant_cuda( + v_transposed_permutted, + v_fp8, + vm, + v_scale, + kv_len, + scale_max, + _tensor_layout, + ) + return v_fp8, v_scale, vm + else: + ops.scale_fuse_quant_cuda( + v_transposed_permutted, v_fp8, v_scale, kv_len, scale_max, _tensor_layout + ) + return v_fp8, v_scale, None diff --git a/build/torch29-cxx11-cu126-x86_64-linux/quant_per_thread.py b/build/torch29-cxx11-cu126-x86_64-linux/quant_per_thread.py new file mode 100644 index 0000000000000000000000000000000000000000..ab81f57c3e6dd9df89946ba54c4e2c3844c94d34 --- /dev/null +++ b/build/torch29-cxx11-cu126-x86_64-linux/quant_per_thread.py @@ -0,0 +1,204 @@ +""" +Copyright (c) 2024 by SageAttention team. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +""" + +import torch +import triton +import triton.language as tl + +@triton.jit +def quant_query_per_thread_int8_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 8 + off_tld = tl.program_id(0) % 8 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + offs_n = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld + offs_k = tl.arange(0, C) + + input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 8 + off_tld + + x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + x = x.to(tl.float32) + scale = tl.max(tl.abs(x)) / 127. + 0.0000001 + x_int8 = x / scale + x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + x_int8 = x_int8.to(tl.int8) + tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + tl.store(scale_ptrs, scale) + +@triton.jit +def quant_key_per_thread_int8_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 4 + off_tld = tl.program_id(0) % 4 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + # offs_n = off_blk * BLK + tl.cat(tl.arange(0, BLK // 8) * 8, tl.arange(0, BLK // 8) * 8 + 1, True) + off_tld * 2 + # offs_k = tl.arange(0, C) + + # input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + # output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + # scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 4 + off_tld + + # x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + # x = x.to(tl.float32) + # scale = tl.max(tl.abs(x)) / 127. + 0.0000001 + # x_int8 = x / scale + # x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + # x_int8 = x_int8.to(tl.int8) + # tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + # tl.store(scale_ptrs, scale) + + offs_n0 = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld * 2 + offs_n1 = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld * 2 + 1 + offs_k = tl.arange(0, C) + + input_ptrs0 = Input + off_b * stride_iz + off_h * stride_ih + offs_n0[:, None] * stride_in + offs_k[None, :] + input_ptrs1 = Input + off_b * stride_iz + off_h * stride_ih + offs_n1[:, None] * stride_in + offs_k[None, :] + output_ptrs0 = Output + off_b * stride_oz + off_h * stride_oh + offs_n0[:, None] * stride_on + offs_k[None, :] + output_ptrs1 = Output + off_b * stride_oz + off_h * stride_oh + offs_n1[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 4 + off_tld + + x0 = tl.load(input_ptrs0, mask=offs_n0[:, None] < L) + x1 = tl.load(input_ptrs1, mask=offs_n1[:, None] < L) + x0 = x0.to(tl.float32) + x1 = x1.to(tl.float32) + scale = max(tl.max(tl.abs(x0)), tl.max(tl.abs(x1))) / 127. + 0.0000001 + x0_int8 = x0 / scale + x1_int8 = x1 / scale + x0_int8 += 0.5 * tl.where(x0_int8 >= 0, 1, -1) + x1_int8 += 0.5 * tl.where(x1_int8 >= 0, 1, -1) + x0_int8 = x0_int8.to(tl.int8) + x1_int8 = x1_int8.to(tl.int8) + tl.store(output_ptrs0, x0_int8, mask=offs_n0[:, None] < L) + tl.store(output_ptrs1, x1_int8, mask=offs_n1[:, None] < L) + tl.store(scale_ptrs, scale) + +@triton.jit +def quant_query_per_thread_int4_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 8 + off_tld = tl.program_id(0) % 8 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + offs_n = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld + offs_k = tl.arange(0, C) + + input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 8 + off_tld + + x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + x = x.to(tl.float32) + scale = tl.max(tl.abs(x)) / 7. + 0.0000001 + x_int8 = x / scale + x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + x_int8 = x_int8.to(tl.int8) + tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + tl.store(scale_ptrs, scale) + +@triton.jit +def quant_key_per_thread_int4_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 4 + off_tld = tl.program_id(0) % 4 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + offs_n = off_blk * BLK + tl.cat(tl.arange(0, BLK // 8) * 8, tl.arange(0, BLK // 8) * 8 + 1, True) + off_tld * 2 + offs_k = tl.arange(0, C) + + input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 4 + off_tld + + x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + x = x.to(tl.float32) + scale = tl.max(tl.abs(x)) / 7. + 0.0000001 + x_int8 = x / scale + x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + x_int8 = x_int8.to(tl.int8) + tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + tl.store(scale_ptrs, scale) + +def per_thread_int8(q, k, km=None, BLKQ=128, WARPQ=32, BLKK=64, WARPK=64, sm_scale=None, tensor_layout="HND"): + q_int8 = torch.empty(q.shape, dtype=torch.int8, device=q.device) + k_int8 = torch.empty(k.shape, dtype=torch.int8, device=k.device) + + if km is not None: + k = k - km + + if tensor_layout == "HND": + b, h_qo, qo_len, head_dim = q.shape + _, h_kv, kv_len, _ = k.shape + + stride_bz_q, stride_h_q, stride_seq_q = q.stride(0), q.stride(1), q.stride(2) + stride_bz_qo, stride_h_qo, stride_seq_qo = q_int8.stride(0), q_int8.stride(1), q_int8.stride(2) + stride_bz_k, stride_h_k, stride_seq_k = k.stride(0), k.stride(1), k.stride(2) + stride_bz_ko, stride_h_ko, stride_seq_ko = k_int8.stride(0), k_int8.stride(1), k_int8.stride(2) + elif tensor_layout == "NHD": + b, qo_len, h_qo, head_dim = q.shape + _, kv_len, h_kv, _ = k.shape + + stride_bz_q, stride_h_q, stride_seq_q = q.stride(0), q.stride(2), q.stride(1) + stride_bz_qo, stride_h_qo, stride_seq_qo = q_int8.stride(0), q_int8.stride(2), q_int8.stride(1) + stride_bz_k, stride_h_k, stride_seq_k = k.stride(0), k.stride(2), k.stride(1) + stride_bz_ko, stride_h_ko, stride_seq_ko = k_int8.stride(0), k_int8.stride(2), k_int8.stride(1) + else: + raise ValueError(f"Unknown tensor layout: {tensor_layout}") + + q_scale = torch.empty((b, h_qo, (qo_len + BLKQ - 1) // BLKQ * (BLKQ // WARPQ) * 8), device=q.device, dtype=torch.float32) + k_scale = torch.empty((b, h_kv, (kv_len + BLKK - 1) // BLKK * (BLKK // WARPK) * 4), device=q.device, dtype=torch.float32) + + if sm_scale is None: + sm_scale = head_dim**-0.5 + + grid = ((qo_len + BLKQ - 1) // BLKQ * (BLKQ // WARPQ) * 8, h_qo, b) + quant_query_per_thread_int8_kernel[grid]( + q, q_int8, q_scale, qo_len, + stride_bz_q, stride_h_q, stride_seq_q, + stride_bz_qo, stride_h_qo, stride_seq_qo, + q_scale.stride(0), q_scale.stride(1), + C=head_dim, BLK=WARPQ + ) + + grid = ((kv_len + BLKK - 1) // BLKK * (BLKK // WARPK) * 4, h_kv, b) + quant_key_per_thread_int8_kernel[grid]( + k, k_int8, k_scale, kv_len, + stride_bz_k, stride_h_k, stride_seq_k, + stride_bz_ko, stride_h_ko, stride_seq_ko, + k_scale.stride(0), k_scale.stride(1), + C=head_dim, BLK=WARPK + ) + + return q_int8, q_scale, k_int8, k_scale \ No newline at end of file diff --git a/build/torch29-cxx11-cu126-x86_64-linux/sage_attention/__init__.py b/build/torch29-cxx11-cu126-x86_64-linux/sage_attention/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..03dbc1afe1cf156661a2b1b22003cd5f599a0309 --- /dev/null +++ b/build/torch29-cxx11-cu126-x86_64-linux/sage_attention/__init__.py @@ -0,0 +1,26 @@ +import ctypes +import sys + +import importlib +from pathlib import Path +from types import ModuleType + +def _import_from_path(file_path: Path) -> ModuleType: + # We cannot use the module name as-is, after adding it to `sys.modules`, + # it would also be used for other imports. So, we make a module name that + # depends on the path for it to be unique using the hex-encoded hash of + # the path. + path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value) + module_name = path_hash + spec = importlib.util.spec_from_file_location(module_name, file_path) + if spec is None: + raise ImportError(f"Cannot load spec for {module_name} from {file_path}") + module = importlib.util.module_from_spec(spec) + if module is None: + raise ImportError(f"Cannot load module {module_name} from spec") + sys.modules[module_name] = module + spec.loader.exec_module(module) # type: ignore + return module + + +globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py"))) diff --git a/build/torch29-cxx11-cu128-x86_64-linux/__init__.py b/build/torch29-cxx11-cu128-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..43b5e35a4154627e6457993372bd46b2cf78b89a --- /dev/null +++ b/build/torch29-cxx11-cu128-x86_64-linux/__init__.py @@ -0,0 +1,11 @@ +from .quant import per_block_int8, per_warp_int8, sub_mean, per_channel_fp8 +from .core import sageattn + + +__all__ = [ + "per_block_int8", + "per_warp_int8", + "sub_mean", + "per_channel_fp8", + "sageattn", +] \ No newline at end of file diff --git a/build/torch29-cxx11-cu128-x86_64-linux/_ops.py b/build/torch29-cxx11-cu128-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..e5a79964f03d3e86fffc66c151aa59b858911e88 --- /dev/null +++ b/build/torch29-cxx11-cu128-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _sage_attention_c08f355_dirty +ops = torch.ops._sage_attention_c08f355_dirty + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_sage_attention_c08f355_dirty::{op_name}" \ No newline at end of file diff --git a/build/torch29-cxx11-cu128-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so b/build/torch29-cxx11-cu128-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so new file mode 100755 index 0000000000000000000000000000000000000000..b6b0a86edd2e99f1d6538b614bafab9c7ce4d95f --- /dev/null +++ b/build/torch29-cxx11-cu128-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc6185d8754dee0263473ccde23c760182ce49c8b6c944e5e3346deedf577409 +size 26559024 diff --git a/build/torch29-cxx11-cu128-x86_64-linux/core.py b/build/torch29-cxx11-cu128-x86_64-linux/core.py new file mode 100644 index 0000000000000000000000000000000000000000..dadc6160869d0697d9752ab068593069d6aa4afa --- /dev/null +++ b/build/torch29-cxx11-cu128-x86_64-linux/core.py @@ -0,0 +1,969 @@ +""" +Copyright (c) 2024 by SageAttention team. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +""" + +import torch +import torch.nn.functional as F + +from ._ops import ops + + +from .quant import per_warp_int8 as per_warp_int8_cuda +from .quant import sub_mean +from .quant import per_channel_fp8 +from .quant_per_thread import per_thread_int8 as per_thread_int8_triton + +from typing import Any, List, Literal, Optional, Tuple, Union +import warnings + + +import subprocess +import re + + +def get_cuda_version(): + try: + output = subprocess.check_output(["nvcc", "--version"]).decode() + match = re.search(r"release (\d+)\.(\d+)", output) + if match: + major, minor = int(match.group(1)), int(match.group(2)) + return major, minor + except Exception as e: + print("Failed to get CUDA version:", e) + return None, None + + +def get_cuda_arch_versions(): + cuda_archs = [] + for i in range(torch.cuda.device_count()): + major, minor = torch.cuda.get_device_capability(i) + cuda_archs.append(f"sm{major}{minor}") + return cuda_archs + + +def sageattn( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + sm_scale: Optional[float] = None, + return_lse: bool = False, + **kwargs: Any, +): + """ + Automatically selects the appropriate implementation of the SageAttention kernel based on the GPU compute capability. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + """ + print("########tensor_layout", tensor_layout) + arch = get_cuda_arch_versions()[q.device.index] + if arch == "sm80": + return sageattn_qk_int8_pv_fp16_cuda( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32", + ) + elif arch == "sm89": + return sageattn_qk_int8_pv_fp8_cuda( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32+fp16", + ) + elif arch == "sm90": + return sageattn_qk_int8_pv_fp8_cuda_sm90( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32+fp32", + ) + elif arch == "sm120": + return sageattn_qk_int8_pv_fp8_cuda( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + qk_quant_gran="per_warp", + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32+fp16", + ) # sm120 has accurate fp32 accumulator for fp8 mma and triton kernel is currently not usable on sm120. + else: + raise ValueError(f"Unsupported CUDA architecture: {arch}") + + +@torch.compiler.disable +def sageattn_qk_int8_pv_fp16_cuda( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + qk_quant_gran: str = "per_thread", + sm_scale: Optional[float] = None, + pv_accum_dtype: str = "fp32", + smooth_k: bool = True, + smooth_v: bool = False, + return_lse: bool = False, + **kwargs: Any, +) -> torch.Tensor: + """ + SageAttention with INT8 quantization for Q and K, FP16 PV with FP16/FP32 accumulation, implemented using CUDA. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + qk_quant_gran : str + The granularity of quantization for Q and K, either "per_warp" or "per_thread". + Default: "per_thread". + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + pv_accum_dtype : str + The dtype of the accumulation of the product of the value tensor and the attention weights, either "fp16", "fp16+fp32" or "fp32". + - "fp16": PV accumulation is done in fully in FP16. This is the fastest option but may lead to numerical instability. `smooth_v` option will increase the accuracy in cases when the value tensor has a large bias (like in CogVideoX-2b). + - "fp32": PV accumulation is done in FP32. This is the most accurate option but may be slower than "fp16" due to CUDA core overhead. + - "fp16+fp32": PV accumulation is done in FP16, but added to a FP32 buffer every few iterations. This offers a balance between speed and accuracy. + Default: "fp32". + + smooth_k : bool + Whether to smooth the key tensor by subtracting the mean along the sequence dimension. + Default: True. + + smooth_v : bool + Whether to smooth the value tensor by subtracting the mean along the sequence dimension. + smooth_v will be ignored if pv_accum_dtype is "fp32" or "fp16+fp32". + Default: False. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + - `smooth_k` will introduce slight overhead but will improve the accuracy under most circumstances. + """ + + dtype = q.dtype + assert q.is_cuda, "Input tensors must be on cuda." + assert dtype in [torch.float16, torch.bfloat16], ( + "Input tensors must be in dtype of torch.float16 or torch.bfloat16" + ) + assert qk_quant_gran in ["per_warp", "per_thread"], ( + "qk_quant_gran must be either 'per_warp' or 'per_thread'." + ) + assert q.device == k.device == v.device, "All tensors must be on the same device." + assert q.dtype == k.dtype == v.dtype, "All tensors must have the same dtype." + + # FIXME(DefTruth): make sage attention work compatible with distributed + # env, for example, xDiT which launch by torchrun. Without this workaround, + # sage attention will run into illegal memory access error after first + # inference step in distributed env for multi gpus inference. This small + # workaround also make sage attention work compatible with torch.compile + # through non-fullgraph compile mode. + torch.cuda.set_device(v.device) + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + _is_caual = 1 if is_causal else 0 + _qk_quant_gran = 3 if qk_quant_gran == "per_thread" else 2 + _return_lse = 1 if return_lse else 0 + + head_dim_og = q.size(-1) + + if head_dim_og < 64: + q = torch.nn.functional.pad(q, (0, 64 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 64 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 64 - head_dim_og)) + elif head_dim_og > 64 and head_dim_og < 128: + q = torch.nn.functional.pad(q, (0, 128 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 128 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 128 - head_dim_og)) + elif head_dim_og > 128: + raise ValueError(f"Unsupported head_dim: {head_dim_og}") + + # assert last dim is contiguous + assert q.stride(-1) == 1 and k.stride(-1) == 1 and v.stride(-1) == 1, ( + "Last dim of qkv must be contiguous." + ) + + if sm_scale is None: + sm_scale = head_dim_og**-0.5 + + seq_dim = 1 if _tensor_layout == 0 else 2 + nh_dim = 2 if _tensor_layout == 0 else 1 + + if smooth_k: + km = k.mean(dim=seq_dim, keepdim=True) + nqheads = q.size(2) + nkheads = k.size(2) + q_per_kv_heads = nqheads // nkheads + if q_per_kv_heads > 1: + # nheads_k => nheads_q + km_broadcast = torch.repeat_interleave(km, q_per_kv_heads, dim=nh_dim) + else: + km_broadcast = km + if return_lse: + if tensor_layout == "NHD": + lse_correction = ( + torch.matmul( + q.transpose(1, 2), km_broadcast.transpose(1, 2).transpose(2, 3) + ) + .squeeze(-1) + .to(torch.float32) + ) + else: + lse_correction = ( + torch.matmul(q, km_broadcast.transpose(2, 3)) + .squeeze(-1) + .to(torch.float32) + ) + else: + km = None + + if qk_quant_gran == "per_warp": + q_int8, q_scale, k_int8, k_scale = per_warp_int8_cuda( + q, + k, + km, + tensor_layout=tensor_layout, + BLKQ=128, + WARPQ=(16 if (q.size(-1) == 128 and pv_accum_dtype == "fp16+fp32") else 32), + BLKK=64, + ) + elif qk_quant_gran == "per_thread": + q_int8, q_scale, k_int8, k_scale = per_thread_int8_triton( + q, + k, + km, + tensor_layout=tensor_layout, + BLKQ=128, + WARPQ=(16 if (q.size(-1) == 128 and pv_accum_dtype == "fp16+fp32") else 32), + BLKK=64, + WARPK=64, + ) + + o = torch.empty(q.size(), dtype=dtype, device=q.device) + + if pv_accum_dtype in ["fp32", "fp16+fp32"] and smooth_v: + warnings.warn(f"pv_accum_dtype is {pv_accum_dtype}, smooth_v will be ignored.") + smooth_v = False + + if pv_accum_dtype == "fp32": + v = v.to(torch.float16) + lse = ops.qk_int8_sv_f16_accum_f32_attn( + q_int8, + k_int8, + v, + o, + q_scale, + k_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + elif pv_accum_dtype == "fp16": + if smooth_v: + smoothed_v, vm = sub_mean(v, tensor_layout=tensor_layout) + lse = ops.qk_int8_sv_f16_accum_f16_fuse_v_mean_attn( + q_int8, + k_int8, + smoothed_v, + o, + q_scale, + k_scale, + vm, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + else: + v = v.to(torch.float16) + lse = ops.qk_int8_sv_f16_accum_f16_attn( + q_int8, + k_int8, + v, + o, + q_scale, + k_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + elif pv_accum_dtype == "fp16+fp32": + v = v.to(torch.float16) + lse = ops.qk_int8_sv_f16_accum_f16_attn_inst_buf( + q_int8, + k_int8, + v, + o, + q_scale, + k_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + else: + raise ValueError(f"Unsupported pv_accum_dtype: {pv_accum_dtype}") + + o = o[..., :head_dim_og] + + if return_lse: + return ( + o, + lse / 1.44269504 + lse_correction * sm_scale + if smooth_k + else lse / 1.44269504, + ) + else: + return o + + +@torch.compiler.disable +def sageattn_qk_int8_pv_fp8_cuda( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + qk_quant_gran: str = "per_thread", + sm_scale: Optional[float] = None, + pv_accum_dtype: str = "fp32+fp16", + smooth_k: bool = True, + smooth_v: bool = False, + return_lse: bool = False, + **kwargs: Any, +) -> torch.Tensor: + """ + SageAttention with INT8 quantization for Q and K, FP8 PV with FP32 accumulation, implemented using CUDA. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + qk_quant_gran : str + The granularity of quantization for Q and K, either "per_warp" or "per_thread". + Default: "per_thread". + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + pv_accum_dtype : str + The dtype of the accumulation of the product of the value tensor and the attention weights, either "fp32" or "fp32+fp32". + - "fp32": PV accumulation is done in fully in FP32. However, due to the hardware issue, there are only 22 valid bits in the FP32 accumulator. + - "fp32+fp32": PV accumulation is done in FP32 (actually FP22), but added to a FP32 buffer every few iterations. This offers a balance between speed and accuracy. + Default: "fp32+fp32". + + smooth_k : bool + Whether to smooth the key tensor by subtracting the mean along the sequence dimension. + Default: True. + + smooth_v : bool + Whether to smooth the value tensor by subtracting the mean along the sequence dimension. + smooth_v will be ignored if pv_accum_dtype is "fp32+fp32". + Default: False. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + - `smooth_k` will introduce slight overhead but will improve the accuracy under most circumstances. + """ + + dtype = q.dtype + assert q.is_cuda, "Input tensors must be on cuda." + assert dtype in [torch.float16, torch.bfloat16], ( + "Input tensors must be in dtype of torch.float16 or torch.bfloat16" + ) + assert qk_quant_gran in ["per_warp", "per_thread"], ( + "qk_quant_gran must be either 'per_warp' or 'per_thread'." + ) + assert q.device == k.device == v.device, "All tensors must be on the same device." + assert q.dtype == k.dtype == v.dtype, "All tensors must have the same dtype." + + # cuda_major_version, cuda_minor_version = get_cuda_version() + # if(cuda_major_version, cuda_minor_version) < (12, 8) and pv_accum_dtype == 'fp32+fp16': + # warnings.warn("cuda version < 12.8, change pv_accum_dtype to 'fp32+fp32'") + # pv_accum_dtype = 'fp32+fp32' + + # FIXME(DefTruth): make sage attention work compatible with distributed + # env, for example, xDiT which launch by torchrun. Without this workaround, + # sage attention will run into illegal memory access error after first + # inference step in distributed env for multi gpus inference. This small + # workaround also make sage attention work compatible with torch.compile + # through non-fullgraph compile mode. + torch.cuda.set_device(v.device) + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + _is_caual = 1 if is_causal else 0 + _qk_quant_gran = 3 if qk_quant_gran == "per_thread" else 2 + _return_lse = 1 if return_lse else 0 + + head_dim_og = q.size(-1) + + if head_dim_og < 64: + q = torch.nn.functional.pad(q, (0, 64 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 64 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 64 - head_dim_og)) + elif head_dim_og > 64 and head_dim_og < 128: + q = torch.nn.functional.pad(q, (0, 128 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 128 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 128 - head_dim_og)) + elif head_dim_og > 128: + raise ValueError(f"Unsupported head_dim: {head_dim_og}") + + # assert last dim is contiguous + assert q.stride(-1) == 1 and k.stride(-1) == 1 and v.stride(-1) == 1, ( + "Last dim of qkv must be contiguous." + ) + + if sm_scale is None: + sm_scale = head_dim_og**-0.5 + + seq_dim = 1 if _tensor_layout == 0 else 2 + nh_dim = 2 if _tensor_layout == 0 else 1 + + if smooth_k: + km = k.mean(dim=seq_dim, keepdim=True) + nqheads = q.size(2) + nkheads = k.size(2) + q_per_kv_heads = nqheads // nkheads + if q_per_kv_heads > 1: + # nheads_k => nheads_q + km_broadcast = torch.repeat_interleave(km, q_per_kv_heads, dim=nh_dim) + else: + km_broadcast = km + if return_lse: + if tensor_layout == "NHD": + lse_correction = ( + torch.matmul( + q.transpose(1, 2), km_broadcast.transpose(1, 2).transpose(2, 3) + ) + .squeeze(-1) + .to(torch.float32) + ) + else: + lse_correction = ( + torch.matmul(q, km_broadcast.transpose(2, 3)) + .squeeze(-1) + .to(torch.float32) + ) + else: + km = None + + if qk_quant_gran == "per_warp": + q_int8, q_scale, k_int8, k_scale = per_warp_int8_cuda( + q, k, km, tensor_layout=tensor_layout, BLKQ=128, WARPQ=32, BLKK=64 + ) + elif qk_quant_gran == "per_thread": + q_int8, q_scale, k_int8, k_scale = per_thread_int8_triton( + q, k, km, tensor_layout=tensor_layout, BLKQ=128, WARPQ=32, BLKK=64, WARPK=64 + ) + + o = torch.empty(q.size(), dtype=dtype, device=q.device) + + if pv_accum_dtype == "fp32+fp32" and smooth_v: + warnings.warn("pv_accum_dtype is 'fp32+fp32', smooth_v will be ignored.") + smooth_v = False + + if pv_accum_dtype == "fp32+fp16" and smooth_v: + warnings.warn("pv_accum_dtype is 'fp32+fp16', smooth_v will be ignored.") + smooth_v = False + + quant_v_scale_max = 448.0 + if pv_accum_dtype == "fp32+fp16": + quant_v_scale_max = 2.25 + + v_fp8, v_scale, vm = per_channel_fp8( + v, tensor_layout=tensor_layout, scale_max=quant_v_scale_max, smooth_v=smooth_v + ) + print("before kernel call") + if pv_accum_dtype == "fp32": + if smooth_v: + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_fuse_v_mean_attn( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + vm, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + else: + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + elif pv_accum_dtype == "fp32+fp32": + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + elif pv_accum_dtype == "fp32+fp16": + lse = ops.qk_int8_sv_f8_accum_f16_fuse_v_scale_attn_inst_buf( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + o = o[..., :head_dim_og] + print("after kernel call") + if return_lse: + return ( + o, + lse / 1.44269504 + lse_correction * sm_scale + if smooth_k + else lse / 1.44269504, + ) + else: + return o + + +@torch.compiler.disable +def sageattn_qk_int8_pv_fp8_cuda_sm90( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + qk_quant_gran: str = "per_thread", + sm_scale: Optional[float] = None, + pv_accum_dtype: str = "fp32+fp32", + smooth_k: bool = True, + return_lse: bool = False, + **kwargs: Any, +) -> torch.Tensor: + """ + SageAttention with INT8 quantization for Q and K, FP8 PV with FP32 accumulation, implemented using CUDA. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + qk_quant_gran : str + The granularity of quantization for Q and K, either "per_warp" or "per_thread". + Default: "per_thread". + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + pv_accum_dtype : str + The dtype of the accumulation of the product of the value tensor and the attention weights, either "fp32" or "fp32+fp32". + - "fp32": PV accumulation is done in fully in FP32. However, due to the hardware issue, there are only 22 valid bits in the FP32 accumulator. + - "fp32+fp32": PV accumulation is done in FP32 (actually FP22), but added to a FP32 buffer every few iterations. This offers a balance between speed and accuracy. + Default: "fp32+fp32". + + smooth_k : bool + Whether to smooth the key tensor by subtracting the mean along the sequence dimension. + Default: True. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + - `smooth_k` will introduce slight overhead but will improve the accuracy under most circumstances. + """ + + dtype = q.dtype + assert q.is_cuda, "Input tensors must be on cuda." + assert dtype in [torch.float16, torch.bfloat16], ( + "Input tensors must be in dtype of torch.float16 or torch.bfloat16" + ) + assert qk_quant_gran in ["per_warp", "per_thread"], ( + "qk_quant_gran must be either 'per_warp' or 'per_thread'." + ) + assert q.device == k.device == v.device, "All tensors must be on the same device." + assert q.dtype == k.dtype == v.dtype, "All tensors must have the same dtype." + + torch.cuda.set_device(v.device) + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + _is_caual = 1 if is_causal else 0 + _qk_quant_gran = 3 if qk_quant_gran == "per_thread" else 2 + _return_lse = 1 if return_lse else 0 + + head_dim_og = q.size(-1) + + if head_dim_og < 64: + q = torch.nn.functional.pad(q, (0, 64 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 64 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 64 - head_dim_og)) + elif head_dim_og > 64 and head_dim_og < 128: + q = torch.nn.functional.pad(q, (0, 128 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 128 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 128 - head_dim_og)) + elif head_dim_og > 128: + raise ValueError(f"Unsupported head_dim: {head_dim_og}") + + # assert last dim is contiguous + assert q.stride(-1) == 1 and k.stride(-1) == 1 and v.stride(-1) == 1, ( + "Last dim of qkv must be contiguous." + ) + + if sm_scale is None: + sm_scale = head_dim_og**-0.5 + + seq_dim = 1 if _tensor_layout == 0 else 2 + nh_dim = 2 if _tensor_layout == 0 else 1 + + if smooth_k: + km = k.mean(dim=seq_dim, keepdim=True) + nqheads = q.size(2) + nkheads = k.size(2) + q_per_kv_heads = nqheads // nkheads + if q_per_kv_heads > 1: + # nheads_k => nheads_q + km_broadcast = torch.repeat_interleave(km, q_per_kv_heads, dim=nh_dim) + else: + km_broadcast = km + if return_lse: + if tensor_layout == "NHD": + lse_correction = ( + torch.matmul( + q.transpose(1, 2), km_broadcast.transpose(1, 2).transpose(2, 3) + ) + .squeeze(-1) + .to(torch.float32) + ) + else: + lse_correction = ( + torch.matmul(q, km_broadcast.transpose(2, 3)) + .squeeze(-1) + .to(torch.float32) + ) + else: + km = None + + if qk_quant_gran == "per_warp": + q_int8, q_scale, k_int8, k_scale = per_warp_int8_cuda( + q, k, km, tensor_layout=tensor_layout, BLKQ=64, WARPQ=16, BLKK=128 + ) + elif qk_quant_gran == "per_thread": + q_int8, q_scale, k_int8, k_scale = per_thread_int8_triton( + q, + k, + km, + tensor_layout=tensor_layout, + BLKQ=64, + WARPQ=16, + BLKK=128, + WARPK=128, + ) + + o = torch.empty(q.size(), dtype=dtype, device=q.device) + + # pad v to multiple of 128 + # TODO: modify per_channel_fp8 kernel to handle this + kv_len = k.size(seq_dim) + v_pad_len = 128 - (kv_len % 128) if kv_len % 128 != 0 else 0 + if v_pad_len > 0: + if tensor_layout == "HND": + v = torch.cat( + [ + v, + torch.zeros( + v.size(0), + v.size(1), + v_pad_len, + v.size(3), + dtype=v.dtype, + device=v.device, + ), + ], + dim=2, + ) + else: + v = torch.cat( + [ + v, + torch.zeros( + v.size(0), + v_pad_len, + v.size(2), + v.size(3), + dtype=v.dtype, + device=v.device, + ), + ], + dim=1, + ) + + v_fp8, v_scale, _ = per_channel_fp8(v, tensor_layout=tensor_layout, smooth_v=False) + + if pv_accum_dtype == "fp32": + raise NotImplementedError("Please use pv_accum_dtype='fp32+fp32' for sm90.") + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + elif pv_accum_dtype == "fp32+fp32": + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf_sm90( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + + o = o[..., :head_dim_og] + + if return_lse: + return ( + o, + lse / 1.44269504 + lse_correction * sm_scale + if smooth_k + else lse / 1.44269504, + ) + else: + return o diff --git a/build/torch29-cxx11-cu128-x86_64-linux/metadata.json b/build/torch29-cxx11-cu128-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..76bafa5f33b6818aa6bb4cab04be811b87519b44 --- /dev/null +++ b/build/torch29-cxx11-cu128-x86_64-linux/metadata.json @@ -0,0 +1 @@ +{"python-depends":[]} \ No newline at end of file diff --git a/build/torch29-cxx11-cu128-x86_64-linux/quant.py b/build/torch29-cxx11-cu128-x86_64-linux/quant.py new file mode 100644 index 0000000000000000000000000000000000000000..2e7a32c6a1e502bee12d0e0564ff2b90f6b00462 --- /dev/null +++ b/build/torch29-cxx11-cu128-x86_64-linux/quant.py @@ -0,0 +1,326 @@ +""" +Copyright (c) 2024 by SageAttention team. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +""" + +import torch +from typing import Optional + +from ._ops import ops + + +def per_block_int8( + q: torch.Tensor, + k: torch.Tensor, + km: Optional[torch.Tensor] = None, + BLKQ: int = 128, + BLKK: int = 64, + sm_scale: Optional[float] = None, + tensor_layout: str = "HND", +): + """ + Quantize the query tensor `q` and the key tensor `k` with per block quantization. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + km : Optional[torch.Tensor] + The mean tensor of `k` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]``. + Should be of the same dtype as `k` if provided. Default is None. + + sm_scale : Optional[float] + The scale factor for the softmax operation. Default is ``head_dim**-0.5``. + It will be multiplied by ``1.44269504`` to work together with the triton attention kernel. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor] + A tuple containing: + - The quantized query tensor. Shape: Same as `q` but with `int8` dtype. + - The scale tensor of the query tensor. Shape: ``[batch_size, num_qo_heads, (qo_len + BLKQ - 1) // BLKQ]`` with `float32` dtype. + - The quantized key tensor. Shape: Same as `k` but with `int8` dtype. + - The scale tensor of the key tensor. Shape: ``[batch_size, num_kv_heads, (kv_len + BLKK - 1) // BLKK]`` with `float32` dtype. + + Note + ---- + - The tensors `q` and `k` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + """ + + q_int8 = torch.empty(q.shape, dtype=torch.int8, device=q.device) + k_int8 = torch.empty(k.shape, dtype=torch.int8, device=k.device) + + if tensor_layout == "HND": + b, h_qo, qo_len, head_dim = q.shape + _, h_kv, kv_len, _ = k.shape + + elif tensor_layout == "NHD": + b, qo_len, h_qo, head_dim = q.shape + _, kv_len, h_kv, _ = k.shape + + else: + raise ValueError(f"Unknown tensor layout: {tensor_layout}") + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + + q_scale = torch.empty( + (b, h_qo, (qo_len + BLKQ - 1) // BLKQ), device=q.device, dtype=torch.float32 + ) + k_scale = torch.empty( + (b, h_kv, (kv_len + BLKK - 1) // BLKK), device=q.device, dtype=torch.float32 + ) + + if sm_scale is None: + sm_scale = head_dim**-0.5 + + sm_scale *= 1.44269504 + + ops.quant_per_block_int8_cuda(q, q_int8, q_scale, sm_scale, BLKQ, _tensor_layout) + if km is not None: + km = km.squeeze(1) if _tensor_layout == 0 else km.squeeze(2) + ops.quant_per_block_int8_fuse_sub_mean_cuda( + k, km, k_int8, k_scale, BLKK, _tensor_layout + ) + else: + # The bound CUDA op expects an sm_scale argument; use 1.0 for K to avoid scaling + ops.quant_per_block_int8_cuda(k, k_int8, k_scale, 1.0, BLKK, _tensor_layout) + + return q_int8, q_scale, k_int8, k_scale + + +def per_warp_int8( + q: torch.Tensor, + k: torch.Tensor, + km: Optional[torch.Tensor] = None, + BLKQ: int = 128, + WARPQ: int = 32, + BLKK: int = 64, + tensor_layout: str = "HND", +): + """ + Quantize the query tensor `q` with per warp quantization and the key tensor `k` with per block quantization. + Warp size of quantizing `q` is 16 or 32, with a block size of 64 or 128. + Block size of quantizing `k` is 64 or 128. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + km : Optional[torch.Tensor] + The mean tensor of `k` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]``. + Should be of the same dtype as `k` if provided. Default is None. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor] + A tuple containing: + - The quantized query tensor. Shape: Same as `q` but with `int8` dtype. + - The scale tensor of the query tensor. Shape: ``[batch_size, num_qo_heads, (qo_len + BLKQ - 1) // BLKQ * (BLKQ // WARPQ)]`` with `float32` dtype. + - The quantized key tensor. Shape: Same as `k` but with `int8` dtype. + - The scale tensor of the key tensor. Shape: ``[batch_size, num_kv_heads, (kv_len + BLKK - 1) // BLKK]`` with `float32` dtype. + + Note + ---- + - The tensors `q` and `k` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + """ + + q_int8 = torch.empty(q.shape, dtype=torch.int8, device=q.device) + k_int8 = torch.empty(k.shape, dtype=torch.int8, device=k.device) + + if tensor_layout == "HND": + b, h_qo, qo_len, head_dim = q.shape + _, h_kv, kv_len, _ = k.shape + + elif tensor_layout == "NHD": + b, qo_len, h_qo, head_dim = q.shape + _, kv_len, h_kv, _ = k.shape + + else: + raise ValueError(f"Unknown tensor layout: {tensor_layout}") + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + + q_scale = torch.empty( + (b, h_qo, ((qo_len + BLKQ - 1) // BLKQ) * (BLKQ // WARPQ)), + device=q.device, + dtype=torch.float32, + ) + k_scale = torch.empty( + (b, h_kv, (kv_len + BLKK - 1) // BLKK), device=q.device, dtype=torch.float32 + ) + + ops.quant_per_warp_int8_cuda(q, q_int8, q_scale, BLKQ, WARPQ, _tensor_layout) + + if km is not None: + km = km.squeeze(1) if _tensor_layout == 0 else km.squeeze(2) + ops.quant_per_block_int8_fuse_sub_mean_cuda( + k, km, k_int8, k_scale, BLKK, _tensor_layout + ) + else: + # The bound CUDA op expects an sm_scale argument; use 1.0 for K to avoid scaling + ops.quant_per_block_int8_cuda(k, k_int8, k_scale, 1.0, BLKK, _tensor_layout) + + return q_int8, q_scale, k_int8, k_scale + + +def sub_mean(v: torch.Tensor, tensor_layout: str = "HND"): + """ + Calculate the mean of the tensor `v` along the sequence length dimension and subtract it from `v`. Result is stored as fp16. + + Parameters + ---------- + v : torch.Tensor + The input tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor] + A tuple containing: + - The tensor `v_smoothed` with the mean subtracted and stored as fp16. Shape: Same as `v` with `float16` dtype. + - The mean tensor of `v` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]`` with dtype same as `v`. + + Note + ---- + - The tensors `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - The returned tensor `v_smoothed` will have dtype ``torch.float16`` regardless of the input dtype. + - The returned mean tensor will have the same dtype as the input tensor. + """ + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + vm = v.mean(dim=1 if _tensor_layout == 0 else 2) + + v_smoothed = torch.empty(v.shape, dtype=torch.float16, device=v.device) + + # subtract mean and store the result as fp16 + ops.sub_mean_cuda(v, vm, v_smoothed, _tensor_layout) + + return v_smoothed, vm + + +def per_channel_fp8( + v: torch.Tensor, + tensor_layout: str = "HND", + scale_max: float = 448.0, + smooth_v: bool = True, +): + """ + Transpose, pad and permute the tensor `v` and quantize it to fp8 with per channel quantization. + `v` is first transposed along the head dimension and the sequence length dimension, then padded to a multiple of 64. + After that, the tensor is permuted along the sequence length dimension by ``[0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15]``. + The quantization is done per channel, with the scale value and smooth factor calculated per channel. + + Parameters + ---------- + v : torch.Tensor + The input tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + scale_max : float + The maximum scale value for the quantization. Default is 448.0 (upper bound of E4M3 data format). + + smooth_v : bool + Whether to smooth the quantized tensor. Default is True. + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor, Optional[torch.Tensor]] + A tuple containing: + - The quantized tensor `v_fp8`. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, head_dim, (kv_len + 63) // 64 * 64]``, with `float8_e4m3fn` dtype. + - If `tensor_layout` is "NHD": ``[batch_size, head_dim, num_kv_heads, (kv_len + 63) // 64 * 64]``, with `float8_e4m3fn` dtype. + - The scale tensor of `v`. Shape: ``[batch_size, num_kv_heads, head_dim]`` with `float32` dtype. + - The mean tensor of `v` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]`` with `float32` dtype. + + Note + ---- + - The tensors `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - The returned mean tensor will be None if `smooth_v` is False. Otherwise it will have dtype ``torch.float32``. + """ + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + + if tensor_layout == "HND": + b, h_kv, kv_len, head_dim = v.shape + padded_len = (kv_len + 63) // 64 * 64 + v_transposed_permutted = torch.empty( + (b, h_kv, head_dim, padded_len), dtype=v.dtype, device=v.device + ) + + elif tensor_layout == "NHD": + b, kv_len, h_kv, head_dim = v.shape + padded_len = (kv_len + 63) // 64 * 64 + v_transposed_permutted = torch.empty( + (b, head_dim, h_kv, padded_len), dtype=v.dtype, device=v.device + ) + + ops.transpose_pad_permute_cuda(v, v_transposed_permutted, _tensor_layout) + + v_fp8 = torch.empty( + v_transposed_permutted.shape, dtype=torch.float8_e4m3fn, device=v.device + ) + + v_scale = torch.empty((b, h_kv, head_dim), dtype=torch.float32, device=v.device) + vm = torch.empty((b, h_kv, head_dim), dtype=torch.float32, device=v.device) + + if smooth_v: + ops.mean_scale_fuse_quant_cuda( + v_transposed_permutted, + v_fp8, + vm, + v_scale, + kv_len, + scale_max, + _tensor_layout, + ) + return v_fp8, v_scale, vm + else: + ops.scale_fuse_quant_cuda( + v_transposed_permutted, v_fp8, v_scale, kv_len, scale_max, _tensor_layout + ) + return v_fp8, v_scale, None diff --git a/build/torch29-cxx11-cu128-x86_64-linux/quant_per_thread.py b/build/torch29-cxx11-cu128-x86_64-linux/quant_per_thread.py new file mode 100644 index 0000000000000000000000000000000000000000..ab81f57c3e6dd9df89946ba54c4e2c3844c94d34 --- /dev/null +++ b/build/torch29-cxx11-cu128-x86_64-linux/quant_per_thread.py @@ -0,0 +1,204 @@ +""" +Copyright (c) 2024 by SageAttention team. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +""" + +import torch +import triton +import triton.language as tl + +@triton.jit +def quant_query_per_thread_int8_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 8 + off_tld = tl.program_id(0) % 8 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + offs_n = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld + offs_k = tl.arange(0, C) + + input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 8 + off_tld + + x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + x = x.to(tl.float32) + scale = tl.max(tl.abs(x)) / 127. + 0.0000001 + x_int8 = x / scale + x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + x_int8 = x_int8.to(tl.int8) + tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + tl.store(scale_ptrs, scale) + +@triton.jit +def quant_key_per_thread_int8_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 4 + off_tld = tl.program_id(0) % 4 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + # offs_n = off_blk * BLK + tl.cat(tl.arange(0, BLK // 8) * 8, tl.arange(0, BLK // 8) * 8 + 1, True) + off_tld * 2 + # offs_k = tl.arange(0, C) + + # input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + # output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + # scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 4 + off_tld + + # x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + # x = x.to(tl.float32) + # scale = tl.max(tl.abs(x)) / 127. + 0.0000001 + # x_int8 = x / scale + # x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + # x_int8 = x_int8.to(tl.int8) + # tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + # tl.store(scale_ptrs, scale) + + offs_n0 = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld * 2 + offs_n1 = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld * 2 + 1 + offs_k = tl.arange(0, C) + + input_ptrs0 = Input + off_b * stride_iz + off_h * stride_ih + offs_n0[:, None] * stride_in + offs_k[None, :] + input_ptrs1 = Input + off_b * stride_iz + off_h * stride_ih + offs_n1[:, None] * stride_in + offs_k[None, :] + output_ptrs0 = Output + off_b * stride_oz + off_h * stride_oh + offs_n0[:, None] * stride_on + offs_k[None, :] + output_ptrs1 = Output + off_b * stride_oz + off_h * stride_oh + offs_n1[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 4 + off_tld + + x0 = tl.load(input_ptrs0, mask=offs_n0[:, None] < L) + x1 = tl.load(input_ptrs1, mask=offs_n1[:, None] < L) + x0 = x0.to(tl.float32) + x1 = x1.to(tl.float32) + scale = max(tl.max(tl.abs(x0)), tl.max(tl.abs(x1))) / 127. + 0.0000001 + x0_int8 = x0 / scale + x1_int8 = x1 / scale + x0_int8 += 0.5 * tl.where(x0_int8 >= 0, 1, -1) + x1_int8 += 0.5 * tl.where(x1_int8 >= 0, 1, -1) + x0_int8 = x0_int8.to(tl.int8) + x1_int8 = x1_int8.to(tl.int8) + tl.store(output_ptrs0, x0_int8, mask=offs_n0[:, None] < L) + tl.store(output_ptrs1, x1_int8, mask=offs_n1[:, None] < L) + tl.store(scale_ptrs, scale) + +@triton.jit +def quant_query_per_thread_int4_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 8 + off_tld = tl.program_id(0) % 8 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + offs_n = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld + offs_k = tl.arange(0, C) + + input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 8 + off_tld + + x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + x = x.to(tl.float32) + scale = tl.max(tl.abs(x)) / 7. + 0.0000001 + x_int8 = x / scale + x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + x_int8 = x_int8.to(tl.int8) + tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + tl.store(scale_ptrs, scale) + +@triton.jit +def quant_key_per_thread_int4_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 4 + off_tld = tl.program_id(0) % 4 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + offs_n = off_blk * BLK + tl.cat(tl.arange(0, BLK // 8) * 8, tl.arange(0, BLK // 8) * 8 + 1, True) + off_tld * 2 + offs_k = tl.arange(0, C) + + input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 4 + off_tld + + x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + x = x.to(tl.float32) + scale = tl.max(tl.abs(x)) / 7. + 0.0000001 + x_int8 = x / scale + x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + x_int8 = x_int8.to(tl.int8) + tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + tl.store(scale_ptrs, scale) + +def per_thread_int8(q, k, km=None, BLKQ=128, WARPQ=32, BLKK=64, WARPK=64, sm_scale=None, tensor_layout="HND"): + q_int8 = torch.empty(q.shape, dtype=torch.int8, device=q.device) + k_int8 = torch.empty(k.shape, dtype=torch.int8, device=k.device) + + if km is not None: + k = k - km + + if tensor_layout == "HND": + b, h_qo, qo_len, head_dim = q.shape + _, h_kv, kv_len, _ = k.shape + + stride_bz_q, stride_h_q, stride_seq_q = q.stride(0), q.stride(1), q.stride(2) + stride_bz_qo, stride_h_qo, stride_seq_qo = q_int8.stride(0), q_int8.stride(1), q_int8.stride(2) + stride_bz_k, stride_h_k, stride_seq_k = k.stride(0), k.stride(1), k.stride(2) + stride_bz_ko, stride_h_ko, stride_seq_ko = k_int8.stride(0), k_int8.stride(1), k_int8.stride(2) + elif tensor_layout == "NHD": + b, qo_len, h_qo, head_dim = q.shape + _, kv_len, h_kv, _ = k.shape + + stride_bz_q, stride_h_q, stride_seq_q = q.stride(0), q.stride(2), q.stride(1) + stride_bz_qo, stride_h_qo, stride_seq_qo = q_int8.stride(0), q_int8.stride(2), q_int8.stride(1) + stride_bz_k, stride_h_k, stride_seq_k = k.stride(0), k.stride(2), k.stride(1) + stride_bz_ko, stride_h_ko, stride_seq_ko = k_int8.stride(0), k_int8.stride(2), k_int8.stride(1) + else: + raise ValueError(f"Unknown tensor layout: {tensor_layout}") + + q_scale = torch.empty((b, h_qo, (qo_len + BLKQ - 1) // BLKQ * (BLKQ // WARPQ) * 8), device=q.device, dtype=torch.float32) + k_scale = torch.empty((b, h_kv, (kv_len + BLKK - 1) // BLKK * (BLKK // WARPK) * 4), device=q.device, dtype=torch.float32) + + if sm_scale is None: + sm_scale = head_dim**-0.5 + + grid = ((qo_len + BLKQ - 1) // BLKQ * (BLKQ // WARPQ) * 8, h_qo, b) + quant_query_per_thread_int8_kernel[grid]( + q, q_int8, q_scale, qo_len, + stride_bz_q, stride_h_q, stride_seq_q, + stride_bz_qo, stride_h_qo, stride_seq_qo, + q_scale.stride(0), q_scale.stride(1), + C=head_dim, BLK=WARPQ + ) + + grid = ((kv_len + BLKK - 1) // BLKK * (BLKK // WARPK) * 4, h_kv, b) + quant_key_per_thread_int8_kernel[grid]( + k, k_int8, k_scale, kv_len, + stride_bz_k, stride_h_k, stride_seq_k, + stride_bz_ko, stride_h_ko, stride_seq_ko, + k_scale.stride(0), k_scale.stride(1), + C=head_dim, BLK=WARPK + ) + + return q_int8, q_scale, k_int8, k_scale \ No newline at end of file diff --git a/build/torch29-cxx11-cu128-x86_64-linux/sage_attention/__init__.py b/build/torch29-cxx11-cu128-x86_64-linux/sage_attention/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..03dbc1afe1cf156661a2b1b22003cd5f599a0309 --- /dev/null +++ b/build/torch29-cxx11-cu128-x86_64-linux/sage_attention/__init__.py @@ -0,0 +1,26 @@ +import ctypes +import sys + +import importlib +from pathlib import Path +from types import ModuleType + +def _import_from_path(file_path: Path) -> ModuleType: + # We cannot use the module name as-is, after adding it to `sys.modules`, + # it would also be used for other imports. So, we make a module name that + # depends on the path for it to be unique using the hex-encoded hash of + # the path. + path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value) + module_name = path_hash + spec = importlib.util.spec_from_file_location(module_name, file_path) + if spec is None: + raise ImportError(f"Cannot load spec for {module_name} from {file_path}") + module = importlib.util.module_from_spec(spec) + if module is None: + raise ImportError(f"Cannot load module {module_name} from spec") + sys.modules[module_name] = module + spec.loader.exec_module(module) # type: ignore + return module + + +globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py"))) diff --git a/build/torch29-cxx11-cu130-x86_64-linux/__init__.py b/build/torch29-cxx11-cu130-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..43b5e35a4154627e6457993372bd46b2cf78b89a --- /dev/null +++ b/build/torch29-cxx11-cu130-x86_64-linux/__init__.py @@ -0,0 +1,11 @@ +from .quant import per_block_int8, per_warp_int8, sub_mean, per_channel_fp8 +from .core import sageattn + + +__all__ = [ + "per_block_int8", + "per_warp_int8", + "sub_mean", + "per_channel_fp8", + "sageattn", +] \ No newline at end of file diff --git a/build/torch29-cxx11-cu130-x86_64-linux/_ops.py b/build/torch29-cxx11-cu130-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..e5a79964f03d3e86fffc66c151aa59b858911e88 --- /dev/null +++ b/build/torch29-cxx11-cu130-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _sage_attention_c08f355_dirty +ops = torch.ops._sage_attention_c08f355_dirty + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_sage_attention_c08f355_dirty::{op_name}" \ No newline at end of file diff --git a/build/torch29-cxx11-cu130-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so b/build/torch29-cxx11-cu130-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so new file mode 100755 index 0000000000000000000000000000000000000000..3d7b1234f65027627dfef493954860eacef8c618 --- /dev/null +++ b/build/torch29-cxx11-cu130-x86_64-linux/_sage_attention_c08f355_dirty.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:842d8d47ead7226d27a2be80e16d3847725a7436db3709ffd1830d35079d2c9d +size 27230808 diff --git a/build/torch29-cxx11-cu130-x86_64-linux/core.py b/build/torch29-cxx11-cu130-x86_64-linux/core.py new file mode 100644 index 0000000000000000000000000000000000000000..dadc6160869d0697d9752ab068593069d6aa4afa --- /dev/null +++ b/build/torch29-cxx11-cu130-x86_64-linux/core.py @@ -0,0 +1,969 @@ +""" +Copyright (c) 2024 by SageAttention team. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +""" + +import torch +import torch.nn.functional as F + +from ._ops import ops + + +from .quant import per_warp_int8 as per_warp_int8_cuda +from .quant import sub_mean +from .quant import per_channel_fp8 +from .quant_per_thread import per_thread_int8 as per_thread_int8_triton + +from typing import Any, List, Literal, Optional, Tuple, Union +import warnings + + +import subprocess +import re + + +def get_cuda_version(): + try: + output = subprocess.check_output(["nvcc", "--version"]).decode() + match = re.search(r"release (\d+)\.(\d+)", output) + if match: + major, minor = int(match.group(1)), int(match.group(2)) + return major, minor + except Exception as e: + print("Failed to get CUDA version:", e) + return None, None + + +def get_cuda_arch_versions(): + cuda_archs = [] + for i in range(torch.cuda.device_count()): + major, minor = torch.cuda.get_device_capability(i) + cuda_archs.append(f"sm{major}{minor}") + return cuda_archs + + +def sageattn( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + sm_scale: Optional[float] = None, + return_lse: bool = False, + **kwargs: Any, +): + """ + Automatically selects the appropriate implementation of the SageAttention kernel based on the GPU compute capability. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + """ + print("########tensor_layout", tensor_layout) + arch = get_cuda_arch_versions()[q.device.index] + if arch == "sm80": + return sageattn_qk_int8_pv_fp16_cuda( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32", + ) + elif arch == "sm89": + return sageattn_qk_int8_pv_fp8_cuda( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32+fp16", + ) + elif arch == "sm90": + return sageattn_qk_int8_pv_fp8_cuda_sm90( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32+fp32", + ) + elif arch == "sm120": + return sageattn_qk_int8_pv_fp8_cuda( + q, + k, + v, + tensor_layout=tensor_layout, + is_causal=is_causal, + qk_quant_gran="per_warp", + sm_scale=sm_scale, + return_lse=return_lse, + pv_accum_dtype="fp32+fp16", + ) # sm120 has accurate fp32 accumulator for fp8 mma and triton kernel is currently not usable on sm120. + else: + raise ValueError(f"Unsupported CUDA architecture: {arch}") + + +@torch.compiler.disable +def sageattn_qk_int8_pv_fp16_cuda( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + qk_quant_gran: str = "per_thread", + sm_scale: Optional[float] = None, + pv_accum_dtype: str = "fp32", + smooth_k: bool = True, + smooth_v: bool = False, + return_lse: bool = False, + **kwargs: Any, +) -> torch.Tensor: + """ + SageAttention with INT8 quantization for Q and K, FP16 PV with FP16/FP32 accumulation, implemented using CUDA. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + qk_quant_gran : str + The granularity of quantization for Q and K, either "per_warp" or "per_thread". + Default: "per_thread". + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + pv_accum_dtype : str + The dtype of the accumulation of the product of the value tensor and the attention weights, either "fp16", "fp16+fp32" or "fp32". + - "fp16": PV accumulation is done in fully in FP16. This is the fastest option but may lead to numerical instability. `smooth_v` option will increase the accuracy in cases when the value tensor has a large bias (like in CogVideoX-2b). + - "fp32": PV accumulation is done in FP32. This is the most accurate option but may be slower than "fp16" due to CUDA core overhead. + - "fp16+fp32": PV accumulation is done in FP16, but added to a FP32 buffer every few iterations. This offers a balance between speed and accuracy. + Default: "fp32". + + smooth_k : bool + Whether to smooth the key tensor by subtracting the mean along the sequence dimension. + Default: True. + + smooth_v : bool + Whether to smooth the value tensor by subtracting the mean along the sequence dimension. + smooth_v will be ignored if pv_accum_dtype is "fp32" or "fp16+fp32". + Default: False. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + - `smooth_k` will introduce slight overhead but will improve the accuracy under most circumstances. + """ + + dtype = q.dtype + assert q.is_cuda, "Input tensors must be on cuda." + assert dtype in [torch.float16, torch.bfloat16], ( + "Input tensors must be in dtype of torch.float16 or torch.bfloat16" + ) + assert qk_quant_gran in ["per_warp", "per_thread"], ( + "qk_quant_gran must be either 'per_warp' or 'per_thread'." + ) + assert q.device == k.device == v.device, "All tensors must be on the same device." + assert q.dtype == k.dtype == v.dtype, "All tensors must have the same dtype." + + # FIXME(DefTruth): make sage attention work compatible with distributed + # env, for example, xDiT which launch by torchrun. Without this workaround, + # sage attention will run into illegal memory access error after first + # inference step in distributed env for multi gpus inference. This small + # workaround also make sage attention work compatible with torch.compile + # through non-fullgraph compile mode. + torch.cuda.set_device(v.device) + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + _is_caual = 1 if is_causal else 0 + _qk_quant_gran = 3 if qk_quant_gran == "per_thread" else 2 + _return_lse = 1 if return_lse else 0 + + head_dim_og = q.size(-1) + + if head_dim_og < 64: + q = torch.nn.functional.pad(q, (0, 64 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 64 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 64 - head_dim_og)) + elif head_dim_og > 64 and head_dim_og < 128: + q = torch.nn.functional.pad(q, (0, 128 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 128 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 128 - head_dim_og)) + elif head_dim_og > 128: + raise ValueError(f"Unsupported head_dim: {head_dim_og}") + + # assert last dim is contiguous + assert q.stride(-1) == 1 and k.stride(-1) == 1 and v.stride(-1) == 1, ( + "Last dim of qkv must be contiguous." + ) + + if sm_scale is None: + sm_scale = head_dim_og**-0.5 + + seq_dim = 1 if _tensor_layout == 0 else 2 + nh_dim = 2 if _tensor_layout == 0 else 1 + + if smooth_k: + km = k.mean(dim=seq_dim, keepdim=True) + nqheads = q.size(2) + nkheads = k.size(2) + q_per_kv_heads = nqheads // nkheads + if q_per_kv_heads > 1: + # nheads_k => nheads_q + km_broadcast = torch.repeat_interleave(km, q_per_kv_heads, dim=nh_dim) + else: + km_broadcast = km + if return_lse: + if tensor_layout == "NHD": + lse_correction = ( + torch.matmul( + q.transpose(1, 2), km_broadcast.transpose(1, 2).transpose(2, 3) + ) + .squeeze(-1) + .to(torch.float32) + ) + else: + lse_correction = ( + torch.matmul(q, km_broadcast.transpose(2, 3)) + .squeeze(-1) + .to(torch.float32) + ) + else: + km = None + + if qk_quant_gran == "per_warp": + q_int8, q_scale, k_int8, k_scale = per_warp_int8_cuda( + q, + k, + km, + tensor_layout=tensor_layout, + BLKQ=128, + WARPQ=(16 if (q.size(-1) == 128 and pv_accum_dtype == "fp16+fp32") else 32), + BLKK=64, + ) + elif qk_quant_gran == "per_thread": + q_int8, q_scale, k_int8, k_scale = per_thread_int8_triton( + q, + k, + km, + tensor_layout=tensor_layout, + BLKQ=128, + WARPQ=(16 if (q.size(-1) == 128 and pv_accum_dtype == "fp16+fp32") else 32), + BLKK=64, + WARPK=64, + ) + + o = torch.empty(q.size(), dtype=dtype, device=q.device) + + if pv_accum_dtype in ["fp32", "fp16+fp32"] and smooth_v: + warnings.warn(f"pv_accum_dtype is {pv_accum_dtype}, smooth_v will be ignored.") + smooth_v = False + + if pv_accum_dtype == "fp32": + v = v.to(torch.float16) + lse = ops.qk_int8_sv_f16_accum_f32_attn( + q_int8, + k_int8, + v, + o, + q_scale, + k_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + elif pv_accum_dtype == "fp16": + if smooth_v: + smoothed_v, vm = sub_mean(v, tensor_layout=tensor_layout) + lse = ops.qk_int8_sv_f16_accum_f16_fuse_v_mean_attn( + q_int8, + k_int8, + smoothed_v, + o, + q_scale, + k_scale, + vm, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + else: + v = v.to(torch.float16) + lse = ops.qk_int8_sv_f16_accum_f16_attn( + q_int8, + k_int8, + v, + o, + q_scale, + k_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + elif pv_accum_dtype == "fp16+fp32": + v = v.to(torch.float16) + lse = ops.qk_int8_sv_f16_accum_f16_attn_inst_buf( + q_int8, + k_int8, + v, + o, + q_scale, + k_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + else: + raise ValueError(f"Unsupported pv_accum_dtype: {pv_accum_dtype}") + + o = o[..., :head_dim_og] + + if return_lse: + return ( + o, + lse / 1.44269504 + lse_correction * sm_scale + if smooth_k + else lse / 1.44269504, + ) + else: + return o + + +@torch.compiler.disable +def sageattn_qk_int8_pv_fp8_cuda( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + qk_quant_gran: str = "per_thread", + sm_scale: Optional[float] = None, + pv_accum_dtype: str = "fp32+fp16", + smooth_k: bool = True, + smooth_v: bool = False, + return_lse: bool = False, + **kwargs: Any, +) -> torch.Tensor: + """ + SageAttention with INT8 quantization for Q and K, FP8 PV with FP32 accumulation, implemented using CUDA. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + qk_quant_gran : str + The granularity of quantization for Q and K, either "per_warp" or "per_thread". + Default: "per_thread". + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + pv_accum_dtype : str + The dtype of the accumulation of the product of the value tensor and the attention weights, either "fp32" or "fp32+fp32". + - "fp32": PV accumulation is done in fully in FP32. However, due to the hardware issue, there are only 22 valid bits in the FP32 accumulator. + - "fp32+fp32": PV accumulation is done in FP32 (actually FP22), but added to a FP32 buffer every few iterations. This offers a balance between speed and accuracy. + Default: "fp32+fp32". + + smooth_k : bool + Whether to smooth the key tensor by subtracting the mean along the sequence dimension. + Default: True. + + smooth_v : bool + Whether to smooth the value tensor by subtracting the mean along the sequence dimension. + smooth_v will be ignored if pv_accum_dtype is "fp32+fp32". + Default: False. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + - `smooth_k` will introduce slight overhead but will improve the accuracy under most circumstances. + """ + + dtype = q.dtype + assert q.is_cuda, "Input tensors must be on cuda." + assert dtype in [torch.float16, torch.bfloat16], ( + "Input tensors must be in dtype of torch.float16 or torch.bfloat16" + ) + assert qk_quant_gran in ["per_warp", "per_thread"], ( + "qk_quant_gran must be either 'per_warp' or 'per_thread'." + ) + assert q.device == k.device == v.device, "All tensors must be on the same device." + assert q.dtype == k.dtype == v.dtype, "All tensors must have the same dtype." + + # cuda_major_version, cuda_minor_version = get_cuda_version() + # if(cuda_major_version, cuda_minor_version) < (12, 8) and pv_accum_dtype == 'fp32+fp16': + # warnings.warn("cuda version < 12.8, change pv_accum_dtype to 'fp32+fp32'") + # pv_accum_dtype = 'fp32+fp32' + + # FIXME(DefTruth): make sage attention work compatible with distributed + # env, for example, xDiT which launch by torchrun. Without this workaround, + # sage attention will run into illegal memory access error after first + # inference step in distributed env for multi gpus inference. This small + # workaround also make sage attention work compatible with torch.compile + # through non-fullgraph compile mode. + torch.cuda.set_device(v.device) + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + _is_caual = 1 if is_causal else 0 + _qk_quant_gran = 3 if qk_quant_gran == "per_thread" else 2 + _return_lse = 1 if return_lse else 0 + + head_dim_og = q.size(-1) + + if head_dim_og < 64: + q = torch.nn.functional.pad(q, (0, 64 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 64 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 64 - head_dim_og)) + elif head_dim_og > 64 and head_dim_og < 128: + q = torch.nn.functional.pad(q, (0, 128 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 128 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 128 - head_dim_og)) + elif head_dim_og > 128: + raise ValueError(f"Unsupported head_dim: {head_dim_og}") + + # assert last dim is contiguous + assert q.stride(-1) == 1 and k.stride(-1) == 1 and v.stride(-1) == 1, ( + "Last dim of qkv must be contiguous." + ) + + if sm_scale is None: + sm_scale = head_dim_og**-0.5 + + seq_dim = 1 if _tensor_layout == 0 else 2 + nh_dim = 2 if _tensor_layout == 0 else 1 + + if smooth_k: + km = k.mean(dim=seq_dim, keepdim=True) + nqheads = q.size(2) + nkheads = k.size(2) + q_per_kv_heads = nqheads // nkheads + if q_per_kv_heads > 1: + # nheads_k => nheads_q + km_broadcast = torch.repeat_interleave(km, q_per_kv_heads, dim=nh_dim) + else: + km_broadcast = km + if return_lse: + if tensor_layout == "NHD": + lse_correction = ( + torch.matmul( + q.transpose(1, 2), km_broadcast.transpose(1, 2).transpose(2, 3) + ) + .squeeze(-1) + .to(torch.float32) + ) + else: + lse_correction = ( + torch.matmul(q, km_broadcast.transpose(2, 3)) + .squeeze(-1) + .to(torch.float32) + ) + else: + km = None + + if qk_quant_gran == "per_warp": + q_int8, q_scale, k_int8, k_scale = per_warp_int8_cuda( + q, k, km, tensor_layout=tensor_layout, BLKQ=128, WARPQ=32, BLKK=64 + ) + elif qk_quant_gran == "per_thread": + q_int8, q_scale, k_int8, k_scale = per_thread_int8_triton( + q, k, km, tensor_layout=tensor_layout, BLKQ=128, WARPQ=32, BLKK=64, WARPK=64 + ) + + o = torch.empty(q.size(), dtype=dtype, device=q.device) + + if pv_accum_dtype == "fp32+fp32" and smooth_v: + warnings.warn("pv_accum_dtype is 'fp32+fp32', smooth_v will be ignored.") + smooth_v = False + + if pv_accum_dtype == "fp32+fp16" and smooth_v: + warnings.warn("pv_accum_dtype is 'fp32+fp16', smooth_v will be ignored.") + smooth_v = False + + quant_v_scale_max = 448.0 + if pv_accum_dtype == "fp32+fp16": + quant_v_scale_max = 2.25 + + v_fp8, v_scale, vm = per_channel_fp8( + v, tensor_layout=tensor_layout, scale_max=quant_v_scale_max, smooth_v=smooth_v + ) + print("before kernel call") + if pv_accum_dtype == "fp32": + if smooth_v: + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_fuse_v_mean_attn( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + vm, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + else: + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + elif pv_accum_dtype == "fp32+fp32": + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + elif pv_accum_dtype == "fp32+fp16": + lse = ops.qk_int8_sv_f8_accum_f16_fuse_v_scale_attn_inst_buf( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + torch.cuda.synchronize() + o = o[..., :head_dim_og] + print("after kernel call") + if return_lse: + return ( + o, + lse / 1.44269504 + lse_correction * sm_scale + if smooth_k + else lse / 1.44269504, + ) + else: + return o + + +@torch.compiler.disable +def sageattn_qk_int8_pv_fp8_cuda_sm90( + q: torch.Tensor, + k: torch.Tensor, + v: torch.Tensor, + tensor_layout: str = "HND", + is_causal: bool = False, + qk_quant_gran: str = "per_thread", + sm_scale: Optional[float] = None, + pv_accum_dtype: str = "fp32+fp32", + smooth_k: bool = True, + return_lse: bool = False, + **kwargs: Any, +) -> torch.Tensor: + """ + SageAttention with INT8 quantization for Q and K, FP8 PV with FP32 accumulation, implemented using CUDA. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + v : torch.Tensor + The value tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + is_causal : bool + Whether to apply causal mask to the attention matrix. Only applicable when qo_len == kv_len. + Default: False. + + qk_quant_gran : str + The granularity of quantization for Q and K, either "per_warp" or "per_thread". + Default: "per_thread". + + sm_scale : Optional[float] + The scale used in softmax, if not provided, will be set to ``1.0 / sqrt(head_dim)``. + + pv_accum_dtype : str + The dtype of the accumulation of the product of the value tensor and the attention weights, either "fp32" or "fp32+fp32". + - "fp32": PV accumulation is done in fully in FP32. However, due to the hardware issue, there are only 22 valid bits in the FP32 accumulator. + - "fp32+fp32": PV accumulation is done in FP32 (actually FP22), but added to a FP32 buffer every few iterations. This offers a balance between speed and accuracy. + Default: "fp32+fp32". + + smooth_k : bool + Whether to smooth the key tensor by subtracting the mean along the sequence dimension. + Default: True. + + return_lse : bool + Whether to return the log sum of the exponentiated attention weights. Used for cases like Ring Attention. + Default: False. + + Returns + ------- + torch.Tensor + The output tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + torch.Tensor + The logsumexp of each row of the matrix QK^T * scaling (e.g., log of the softmax normalization factor). + Shape: ``[batch_size, num_qo_heads, qo_len]``. + Only returned if `return_lse` is True. + + Note + ---- + - ``num_qo_heads`` must be divisible by ``num_kv_heads``. + - The tensors `q`, `k`, and `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - All tensors must be on the same cuda device. + - `smooth_k` will introduce slight overhead but will improve the accuracy under most circumstances. + """ + + dtype = q.dtype + assert q.is_cuda, "Input tensors must be on cuda." + assert dtype in [torch.float16, torch.bfloat16], ( + "Input tensors must be in dtype of torch.float16 or torch.bfloat16" + ) + assert qk_quant_gran in ["per_warp", "per_thread"], ( + "qk_quant_gran must be either 'per_warp' or 'per_thread'." + ) + assert q.device == k.device == v.device, "All tensors must be on the same device." + assert q.dtype == k.dtype == v.dtype, "All tensors must have the same dtype." + + torch.cuda.set_device(v.device) + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + _is_caual = 1 if is_causal else 0 + _qk_quant_gran = 3 if qk_quant_gran == "per_thread" else 2 + _return_lse = 1 if return_lse else 0 + + head_dim_og = q.size(-1) + + if head_dim_og < 64: + q = torch.nn.functional.pad(q, (0, 64 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 64 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 64 - head_dim_og)) + elif head_dim_og > 64 and head_dim_og < 128: + q = torch.nn.functional.pad(q, (0, 128 - head_dim_og)) + k = torch.nn.functional.pad(k, (0, 128 - head_dim_og)) + v = torch.nn.functional.pad(v, (0, 128 - head_dim_og)) + elif head_dim_og > 128: + raise ValueError(f"Unsupported head_dim: {head_dim_og}") + + # assert last dim is contiguous + assert q.stride(-1) == 1 and k.stride(-1) == 1 and v.stride(-1) == 1, ( + "Last dim of qkv must be contiguous." + ) + + if sm_scale is None: + sm_scale = head_dim_og**-0.5 + + seq_dim = 1 if _tensor_layout == 0 else 2 + nh_dim = 2 if _tensor_layout == 0 else 1 + + if smooth_k: + km = k.mean(dim=seq_dim, keepdim=True) + nqheads = q.size(2) + nkheads = k.size(2) + q_per_kv_heads = nqheads // nkheads + if q_per_kv_heads > 1: + # nheads_k => nheads_q + km_broadcast = torch.repeat_interleave(km, q_per_kv_heads, dim=nh_dim) + else: + km_broadcast = km + if return_lse: + if tensor_layout == "NHD": + lse_correction = ( + torch.matmul( + q.transpose(1, 2), km_broadcast.transpose(1, 2).transpose(2, 3) + ) + .squeeze(-1) + .to(torch.float32) + ) + else: + lse_correction = ( + torch.matmul(q, km_broadcast.transpose(2, 3)) + .squeeze(-1) + .to(torch.float32) + ) + else: + km = None + + if qk_quant_gran == "per_warp": + q_int8, q_scale, k_int8, k_scale = per_warp_int8_cuda( + q, k, km, tensor_layout=tensor_layout, BLKQ=64, WARPQ=16, BLKK=128 + ) + elif qk_quant_gran == "per_thread": + q_int8, q_scale, k_int8, k_scale = per_thread_int8_triton( + q, + k, + km, + tensor_layout=tensor_layout, + BLKQ=64, + WARPQ=16, + BLKK=128, + WARPK=128, + ) + + o = torch.empty(q.size(), dtype=dtype, device=q.device) + + # pad v to multiple of 128 + # TODO: modify per_channel_fp8 kernel to handle this + kv_len = k.size(seq_dim) + v_pad_len = 128 - (kv_len % 128) if kv_len % 128 != 0 else 0 + if v_pad_len > 0: + if tensor_layout == "HND": + v = torch.cat( + [ + v, + torch.zeros( + v.size(0), + v.size(1), + v_pad_len, + v.size(3), + dtype=v.dtype, + device=v.device, + ), + ], + dim=2, + ) + else: + v = torch.cat( + [ + v, + torch.zeros( + v.size(0), + v_pad_len, + v.size(2), + v.size(3), + dtype=v.dtype, + device=v.device, + ), + ], + dim=1, + ) + + v_fp8, v_scale, _ = per_channel_fp8(v, tensor_layout=tensor_layout, smooth_v=False) + + if pv_accum_dtype == "fp32": + raise NotImplementedError("Please use pv_accum_dtype='fp32+fp32' for sm90.") + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + elif pv_accum_dtype == "fp32+fp32": + lse = ops.qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf_sm90( + q_int8, + k_int8, + v_fp8, + o, + q_scale, + k_scale, + v_scale, + _tensor_layout, + _is_caual, + _qk_quant_gran, + sm_scale, + _return_lse, + ) + + o = o[..., :head_dim_og] + + if return_lse: + return ( + o, + lse / 1.44269504 + lse_correction * sm_scale + if smooth_k + else lse / 1.44269504, + ) + else: + return o diff --git a/build/torch29-cxx11-cu130-x86_64-linux/metadata.json b/build/torch29-cxx11-cu130-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..76bafa5f33b6818aa6bb4cab04be811b87519b44 --- /dev/null +++ b/build/torch29-cxx11-cu130-x86_64-linux/metadata.json @@ -0,0 +1 @@ +{"python-depends":[]} \ No newline at end of file diff --git a/build/torch29-cxx11-cu130-x86_64-linux/quant.py b/build/torch29-cxx11-cu130-x86_64-linux/quant.py new file mode 100644 index 0000000000000000000000000000000000000000..2e7a32c6a1e502bee12d0e0564ff2b90f6b00462 --- /dev/null +++ b/build/torch29-cxx11-cu130-x86_64-linux/quant.py @@ -0,0 +1,326 @@ +""" +Copyright (c) 2024 by SageAttention team. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +""" + +import torch +from typing import Optional + +from ._ops import ops + + +def per_block_int8( + q: torch.Tensor, + k: torch.Tensor, + km: Optional[torch.Tensor] = None, + BLKQ: int = 128, + BLKK: int = 64, + sm_scale: Optional[float] = None, + tensor_layout: str = "HND", +): + """ + Quantize the query tensor `q` and the key tensor `k` with per block quantization. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + km : Optional[torch.Tensor] + The mean tensor of `k` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]``. + Should be of the same dtype as `k` if provided. Default is None. + + sm_scale : Optional[float] + The scale factor for the softmax operation. Default is ``head_dim**-0.5``. + It will be multiplied by ``1.44269504`` to work together with the triton attention kernel. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor] + A tuple containing: + - The quantized query tensor. Shape: Same as `q` but with `int8` dtype. + - The scale tensor of the query tensor. Shape: ``[batch_size, num_qo_heads, (qo_len + BLKQ - 1) // BLKQ]`` with `float32` dtype. + - The quantized key tensor. Shape: Same as `k` but with `int8` dtype. + - The scale tensor of the key tensor. Shape: ``[batch_size, num_kv_heads, (kv_len + BLKK - 1) // BLKK]`` with `float32` dtype. + + Note + ---- + - The tensors `q` and `k` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + """ + + q_int8 = torch.empty(q.shape, dtype=torch.int8, device=q.device) + k_int8 = torch.empty(k.shape, dtype=torch.int8, device=k.device) + + if tensor_layout == "HND": + b, h_qo, qo_len, head_dim = q.shape + _, h_kv, kv_len, _ = k.shape + + elif tensor_layout == "NHD": + b, qo_len, h_qo, head_dim = q.shape + _, kv_len, h_kv, _ = k.shape + + else: + raise ValueError(f"Unknown tensor layout: {tensor_layout}") + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + + q_scale = torch.empty( + (b, h_qo, (qo_len + BLKQ - 1) // BLKQ), device=q.device, dtype=torch.float32 + ) + k_scale = torch.empty( + (b, h_kv, (kv_len + BLKK - 1) // BLKK), device=q.device, dtype=torch.float32 + ) + + if sm_scale is None: + sm_scale = head_dim**-0.5 + + sm_scale *= 1.44269504 + + ops.quant_per_block_int8_cuda(q, q_int8, q_scale, sm_scale, BLKQ, _tensor_layout) + if km is not None: + km = km.squeeze(1) if _tensor_layout == 0 else km.squeeze(2) + ops.quant_per_block_int8_fuse_sub_mean_cuda( + k, km, k_int8, k_scale, BLKK, _tensor_layout + ) + else: + # The bound CUDA op expects an sm_scale argument; use 1.0 for K to avoid scaling + ops.quant_per_block_int8_cuda(k, k_int8, k_scale, 1.0, BLKK, _tensor_layout) + + return q_int8, q_scale, k_int8, k_scale + + +def per_warp_int8( + q: torch.Tensor, + k: torch.Tensor, + km: Optional[torch.Tensor] = None, + BLKQ: int = 128, + WARPQ: int = 32, + BLKK: int = 64, + tensor_layout: str = "HND", +): + """ + Quantize the query tensor `q` with per warp quantization and the key tensor `k` with per block quantization. + Warp size of quantizing `q` is 16 or 32, with a block size of 64 or 128. + Block size of quantizing `k` is 64 or 128. + + Parameters + ---------- + q : torch.Tensor + The query tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_qo_heads, qo_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, qo_len, num_qo_heads, head_dim]``. + + k : torch.Tensor + The key tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + km : Optional[torch.Tensor] + The mean tensor of `k` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]``. + Should be of the same dtype as `k` if provided. Default is None. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor] + A tuple containing: + - The quantized query tensor. Shape: Same as `q` but with `int8` dtype. + - The scale tensor of the query tensor. Shape: ``[batch_size, num_qo_heads, (qo_len + BLKQ - 1) // BLKQ * (BLKQ // WARPQ)]`` with `float32` dtype. + - The quantized key tensor. Shape: Same as `k` but with `int8` dtype. + - The scale tensor of the key tensor. Shape: ``[batch_size, num_kv_heads, (kv_len + BLKK - 1) // BLKK]`` with `float32` dtype. + + Note + ---- + - The tensors `q` and `k` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + """ + + q_int8 = torch.empty(q.shape, dtype=torch.int8, device=q.device) + k_int8 = torch.empty(k.shape, dtype=torch.int8, device=k.device) + + if tensor_layout == "HND": + b, h_qo, qo_len, head_dim = q.shape + _, h_kv, kv_len, _ = k.shape + + elif tensor_layout == "NHD": + b, qo_len, h_qo, head_dim = q.shape + _, kv_len, h_kv, _ = k.shape + + else: + raise ValueError(f"Unknown tensor layout: {tensor_layout}") + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + + q_scale = torch.empty( + (b, h_qo, ((qo_len + BLKQ - 1) // BLKQ) * (BLKQ // WARPQ)), + device=q.device, + dtype=torch.float32, + ) + k_scale = torch.empty( + (b, h_kv, (kv_len + BLKK - 1) // BLKK), device=q.device, dtype=torch.float32 + ) + + ops.quant_per_warp_int8_cuda(q, q_int8, q_scale, BLKQ, WARPQ, _tensor_layout) + + if km is not None: + km = km.squeeze(1) if _tensor_layout == 0 else km.squeeze(2) + ops.quant_per_block_int8_fuse_sub_mean_cuda( + k, km, k_int8, k_scale, BLKK, _tensor_layout + ) + else: + # The bound CUDA op expects an sm_scale argument; use 1.0 for K to avoid scaling + ops.quant_per_block_int8_cuda(k, k_int8, k_scale, 1.0, BLKK, _tensor_layout) + + return q_int8, q_scale, k_int8, k_scale + + +def sub_mean(v: torch.Tensor, tensor_layout: str = "HND"): + """ + Calculate the mean of the tensor `v` along the sequence length dimension and subtract it from `v`. Result is stored as fp16. + + Parameters + ---------- + v : torch.Tensor + The input tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor] + A tuple containing: + - The tensor `v_smoothed` with the mean subtracted and stored as fp16. Shape: Same as `v` with `float16` dtype. + - The mean tensor of `v` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]`` with dtype same as `v`. + + Note + ---- + - The tensors `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - The returned tensor `v_smoothed` will have dtype ``torch.float16`` regardless of the input dtype. + - The returned mean tensor will have the same dtype as the input tensor. + """ + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + vm = v.mean(dim=1 if _tensor_layout == 0 else 2) + + v_smoothed = torch.empty(v.shape, dtype=torch.float16, device=v.device) + + # subtract mean and store the result as fp16 + ops.sub_mean_cuda(v, vm, v_smoothed, _tensor_layout) + + return v_smoothed, vm + + +def per_channel_fp8( + v: torch.Tensor, + tensor_layout: str = "HND", + scale_max: float = 448.0, + smooth_v: bool = True, +): + """ + Transpose, pad and permute the tensor `v` and quantize it to fp8 with per channel quantization. + `v` is first transposed along the head dimension and the sequence length dimension, then padded to a multiple of 64. + After that, the tensor is permuted along the sequence length dimension by ``[0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15]``. + The quantization is done per channel, with the scale value and smooth factor calculated per channel. + + Parameters + ---------- + v : torch.Tensor + The input tensor. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, kv_len, head_dim]``. + - If `tensor_layout` is "NHD": ``[batch_size, kv_len, num_kv_heads, head_dim]``. + + tensor_layout : str + The tensor layout, either "HND" or "NHD". + Default: "HND". + + scale_max : float + The maximum scale value for the quantization. Default is 448.0 (upper bound of E4M3 data format). + + smooth_v : bool + Whether to smooth the quantized tensor. Default is True. + + Returns + ------- + Tuple[torch.Tensor, torch.Tensor, Optional[torch.Tensor]] + A tuple containing: + - The quantized tensor `v_fp8`. Shape: + - If `tensor_layout` is "HND": ``[batch_size, num_kv_heads, head_dim, (kv_len + 63) // 64 * 64]``, with `float8_e4m3fn` dtype. + - If `tensor_layout` is "NHD": ``[batch_size, head_dim, num_kv_heads, (kv_len + 63) // 64 * 64]``, with `float8_e4m3fn` dtype. + - The scale tensor of `v`. Shape: ``[batch_size, num_kv_heads, head_dim]`` with `float32` dtype. + - The mean tensor of `v` along the sequence length dimension. Shape: ``[batch_size, num_kv_heads, head_dim]`` with `float32` dtype. + + Note + ---- + - The tensors `v` must have the dtype ``torch.float16`` or ``torch.bfloat16`` + - The returned mean tensor will be None if `smooth_v` is False. Otherwise it will have dtype ``torch.float32``. + """ + + _tensor_layout = 0 if tensor_layout == "NHD" else 1 + + if tensor_layout == "HND": + b, h_kv, kv_len, head_dim = v.shape + padded_len = (kv_len + 63) // 64 * 64 + v_transposed_permutted = torch.empty( + (b, h_kv, head_dim, padded_len), dtype=v.dtype, device=v.device + ) + + elif tensor_layout == "NHD": + b, kv_len, h_kv, head_dim = v.shape + padded_len = (kv_len + 63) // 64 * 64 + v_transposed_permutted = torch.empty( + (b, head_dim, h_kv, padded_len), dtype=v.dtype, device=v.device + ) + + ops.transpose_pad_permute_cuda(v, v_transposed_permutted, _tensor_layout) + + v_fp8 = torch.empty( + v_transposed_permutted.shape, dtype=torch.float8_e4m3fn, device=v.device + ) + + v_scale = torch.empty((b, h_kv, head_dim), dtype=torch.float32, device=v.device) + vm = torch.empty((b, h_kv, head_dim), dtype=torch.float32, device=v.device) + + if smooth_v: + ops.mean_scale_fuse_quant_cuda( + v_transposed_permutted, + v_fp8, + vm, + v_scale, + kv_len, + scale_max, + _tensor_layout, + ) + return v_fp8, v_scale, vm + else: + ops.scale_fuse_quant_cuda( + v_transposed_permutted, v_fp8, v_scale, kv_len, scale_max, _tensor_layout + ) + return v_fp8, v_scale, None diff --git a/build/torch29-cxx11-cu130-x86_64-linux/quant_per_thread.py b/build/torch29-cxx11-cu130-x86_64-linux/quant_per_thread.py new file mode 100644 index 0000000000000000000000000000000000000000..ab81f57c3e6dd9df89946ba54c4e2c3844c94d34 --- /dev/null +++ b/build/torch29-cxx11-cu130-x86_64-linux/quant_per_thread.py @@ -0,0 +1,204 @@ +""" +Copyright (c) 2024 by SageAttention team. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +""" + +import torch +import triton +import triton.language as tl + +@triton.jit +def quant_query_per_thread_int8_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 8 + off_tld = tl.program_id(0) % 8 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + offs_n = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld + offs_k = tl.arange(0, C) + + input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 8 + off_tld + + x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + x = x.to(tl.float32) + scale = tl.max(tl.abs(x)) / 127. + 0.0000001 + x_int8 = x / scale + x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + x_int8 = x_int8.to(tl.int8) + tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + tl.store(scale_ptrs, scale) + +@triton.jit +def quant_key_per_thread_int8_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 4 + off_tld = tl.program_id(0) % 4 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + # offs_n = off_blk * BLK + tl.cat(tl.arange(0, BLK // 8) * 8, tl.arange(0, BLK // 8) * 8 + 1, True) + off_tld * 2 + # offs_k = tl.arange(0, C) + + # input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + # output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + # scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 4 + off_tld + + # x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + # x = x.to(tl.float32) + # scale = tl.max(tl.abs(x)) / 127. + 0.0000001 + # x_int8 = x / scale + # x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + # x_int8 = x_int8.to(tl.int8) + # tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + # tl.store(scale_ptrs, scale) + + offs_n0 = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld * 2 + offs_n1 = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld * 2 + 1 + offs_k = tl.arange(0, C) + + input_ptrs0 = Input + off_b * stride_iz + off_h * stride_ih + offs_n0[:, None] * stride_in + offs_k[None, :] + input_ptrs1 = Input + off_b * stride_iz + off_h * stride_ih + offs_n1[:, None] * stride_in + offs_k[None, :] + output_ptrs0 = Output + off_b * stride_oz + off_h * stride_oh + offs_n0[:, None] * stride_on + offs_k[None, :] + output_ptrs1 = Output + off_b * stride_oz + off_h * stride_oh + offs_n1[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 4 + off_tld + + x0 = tl.load(input_ptrs0, mask=offs_n0[:, None] < L) + x1 = tl.load(input_ptrs1, mask=offs_n1[:, None] < L) + x0 = x0.to(tl.float32) + x1 = x1.to(tl.float32) + scale = max(tl.max(tl.abs(x0)), tl.max(tl.abs(x1))) / 127. + 0.0000001 + x0_int8 = x0 / scale + x1_int8 = x1 / scale + x0_int8 += 0.5 * tl.where(x0_int8 >= 0, 1, -1) + x1_int8 += 0.5 * tl.where(x1_int8 >= 0, 1, -1) + x0_int8 = x0_int8.to(tl.int8) + x1_int8 = x1_int8.to(tl.int8) + tl.store(output_ptrs0, x0_int8, mask=offs_n0[:, None] < L) + tl.store(output_ptrs1, x1_int8, mask=offs_n1[:, None] < L) + tl.store(scale_ptrs, scale) + +@triton.jit +def quant_query_per_thread_int4_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 8 + off_tld = tl.program_id(0) % 8 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + offs_n = off_blk * BLK + tl.arange(0, BLK // 8) * 8 + off_tld + offs_k = tl.arange(0, C) + + input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 8 + off_tld + + x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + x = x.to(tl.float32) + scale = tl.max(tl.abs(x)) / 7. + 0.0000001 + x_int8 = x / scale + x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + x_int8 = x_int8.to(tl.int8) + tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + tl.store(scale_ptrs, scale) + +@triton.jit +def quant_key_per_thread_int4_kernel(Input, Output, Scale, L, + stride_iz, stride_ih, stride_in, + stride_oz, stride_oh, stride_on, + stride_sz, stride_sh, + C: tl.constexpr, BLK: tl.constexpr): + off_blk = tl.program_id(0) // 4 + off_tld = tl.program_id(0) % 4 + off_h = tl.program_id(1) + off_b = tl.program_id(2) + + offs_n = off_blk * BLK + tl.cat(tl.arange(0, BLK // 8) * 8, tl.arange(0, BLK // 8) * 8 + 1, True) + off_tld * 2 + offs_k = tl.arange(0, C) + + input_ptrs = Input + off_b * stride_iz + off_h * stride_ih + offs_n[:, None] * stride_in + offs_k[None, :] + output_ptrs = Output + off_b * stride_oz + off_h * stride_oh + offs_n[:, None] * stride_on + offs_k[None, :] + scale_ptrs = Scale + off_b * stride_sz + off_h * stride_sh + off_blk * 4 + off_tld + + x = tl.load(input_ptrs, mask=offs_n[:, None] < L) + x = x.to(tl.float32) + scale = tl.max(tl.abs(x)) / 7. + 0.0000001 + x_int8 = x / scale + x_int8 += 0.5 * tl.where(x_int8 >= 0, 1, -1) + x_int8 = x_int8.to(tl.int8) + tl.store(output_ptrs, x_int8, mask=offs_n[:, None] < L) + tl.store(scale_ptrs, scale) + +def per_thread_int8(q, k, km=None, BLKQ=128, WARPQ=32, BLKK=64, WARPK=64, sm_scale=None, tensor_layout="HND"): + q_int8 = torch.empty(q.shape, dtype=torch.int8, device=q.device) + k_int8 = torch.empty(k.shape, dtype=torch.int8, device=k.device) + + if km is not None: + k = k - km + + if tensor_layout == "HND": + b, h_qo, qo_len, head_dim = q.shape + _, h_kv, kv_len, _ = k.shape + + stride_bz_q, stride_h_q, stride_seq_q = q.stride(0), q.stride(1), q.stride(2) + stride_bz_qo, stride_h_qo, stride_seq_qo = q_int8.stride(0), q_int8.stride(1), q_int8.stride(2) + stride_bz_k, stride_h_k, stride_seq_k = k.stride(0), k.stride(1), k.stride(2) + stride_bz_ko, stride_h_ko, stride_seq_ko = k_int8.stride(0), k_int8.stride(1), k_int8.stride(2) + elif tensor_layout == "NHD": + b, qo_len, h_qo, head_dim = q.shape + _, kv_len, h_kv, _ = k.shape + + stride_bz_q, stride_h_q, stride_seq_q = q.stride(0), q.stride(2), q.stride(1) + stride_bz_qo, stride_h_qo, stride_seq_qo = q_int8.stride(0), q_int8.stride(2), q_int8.stride(1) + stride_bz_k, stride_h_k, stride_seq_k = k.stride(0), k.stride(2), k.stride(1) + stride_bz_ko, stride_h_ko, stride_seq_ko = k_int8.stride(0), k_int8.stride(2), k_int8.stride(1) + else: + raise ValueError(f"Unknown tensor layout: {tensor_layout}") + + q_scale = torch.empty((b, h_qo, (qo_len + BLKQ - 1) // BLKQ * (BLKQ // WARPQ) * 8), device=q.device, dtype=torch.float32) + k_scale = torch.empty((b, h_kv, (kv_len + BLKK - 1) // BLKK * (BLKK // WARPK) * 4), device=q.device, dtype=torch.float32) + + if sm_scale is None: + sm_scale = head_dim**-0.5 + + grid = ((qo_len + BLKQ - 1) // BLKQ * (BLKQ // WARPQ) * 8, h_qo, b) + quant_query_per_thread_int8_kernel[grid]( + q, q_int8, q_scale, qo_len, + stride_bz_q, stride_h_q, stride_seq_q, + stride_bz_qo, stride_h_qo, stride_seq_qo, + q_scale.stride(0), q_scale.stride(1), + C=head_dim, BLK=WARPQ + ) + + grid = ((kv_len + BLKK - 1) // BLKK * (BLKK // WARPK) * 4, h_kv, b) + quant_key_per_thread_int8_kernel[grid]( + k, k_int8, k_scale, kv_len, + stride_bz_k, stride_h_k, stride_seq_k, + stride_bz_ko, stride_h_ko, stride_seq_ko, + k_scale.stride(0), k_scale.stride(1), + C=head_dim, BLK=WARPK + ) + + return q_int8, q_scale, k_int8, k_scale \ No newline at end of file diff --git a/build/torch29-cxx11-cu130-x86_64-linux/sage_attention/__init__.py b/build/torch29-cxx11-cu130-x86_64-linux/sage_attention/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..03dbc1afe1cf156661a2b1b22003cd5f599a0309 --- /dev/null +++ b/build/torch29-cxx11-cu130-x86_64-linux/sage_attention/__init__.py @@ -0,0 +1,26 @@ +import ctypes +import sys + +import importlib +from pathlib import Path +from types import ModuleType + +def _import_from_path(file_path: Path) -> ModuleType: + # We cannot use the module name as-is, after adding it to `sys.modules`, + # it would also be used for other imports. So, we make a module name that + # depends on the path for it to be unique using the hex-encoded hash of + # the path. + path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value) + module_name = path_hash + spec = importlib.util.spec_from_file_location(module_name, file_path) + if spec is None: + raise ImportError(f"Cannot load spec for {module_name} from {file_path}") + module = importlib.util.module_from_spec(spec) + if module is None: + raise ImportError(f"Cannot load module {module_name} from spec") + sys.modules[module_name] = module + spec.loader.exec_module(module) # type: ignore + return module + + +globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py"))) diff --git a/flake.lock b/flake.lock index 9adfa6c4875a2c6973af3ce69126543ec51c5a8f..0763581da8dd3bb1bf933b5214f3890ea48f7cea 100644 --- a/flake.lock +++ b/flake.lock @@ -2,26 +2,11 @@ "nodes": { "flake-compat": { "locked": { - "lastModified": 1747046372, - "narHash": "sha256-CIVLLkVgvHYbgI2UpXvIIBJ12HWgX+fjA8Xf8PUmqCY=", + "lastModified": 1761588595, + "narHash": "sha256-XKUZz9zewJNUj46b4AJdiRZJAvSZ0Dqj2BNfXvFlJC4=", "owner": "edolstra", "repo": "flake-compat", - "rev": "9100a0f413b0c601e0533d1d94ffd501ce2e7885", - "type": "github" - }, - "original": { - "owner": "edolstra", - "repo": "flake-compat", - "type": "github" - } - }, - "flake-compat_2": { - "locked": { - "lastModified": 1747046372, - "narHash": "sha256-CIVLLkVgvHYbgI2UpXvIIBJ12HWgX+fjA8Xf8PUmqCY=", - "owner": "edolstra", - "repo": "flake-compat", - "rev": "9100a0f413b0c601e0533d1d94ffd501ce2e7885", + "rev": "f387cd2afec9419c8ee37694406ca490c3f34ee5", "type": "github" }, "original": { @@ -48,82 +33,40 @@ "type": "github" } }, - "flake-utils_2": { - "inputs": { - "systems": "systems_2" - }, - "locked": { - "lastModified": 1731533236, - "narHash": "sha256-l0KFg5HjrsfsO/JpG+r7fRrqm12kzFHyUHqHCVpMMbI=", - "owner": "numtide", - "repo": "flake-utils", - "rev": "11707dc2f618dd54ca8739b309ec4fc024de578b", - "type": "github" - }, - "original": { - "owner": "numtide", - "repo": "flake-utils", - "type": "github" - } - }, - "hf-nix": { - "inputs": { - "flake-compat": "flake-compat_2", - "flake-utils": "flake-utils_2", - "nixpkgs": "nixpkgs" - }, - "locked": { - "lastModified": 1757675377, - "narHash": "sha256-JQKZOI1ZYO4faJnanuoTXziSmqzXe5rEFSGliWDWqWw=", - "owner": "huggingface", - "repo": "hf-nix", - "rev": "faf3354403a7381958d08e826c15fe30f6986a4f", - "type": "github" - }, - "original": { - "owner": "huggingface", - "repo": "hf-nix", - "type": "github" - } - }, "kernel-builder": { "inputs": { "flake-compat": "flake-compat", "flake-utils": "flake-utils", - "hf-nix": "hf-nix", - "nixpkgs": [ - "kernel-builder", - "hf-nix", - "nixpkgs" - ] + "nixpkgs": "nixpkgs" }, "locked": { - "lastModified": 1758789724, - "narHash": "sha256-0EbrJkAx7yTOOjfJQFbk+BLo4MYfsD6JgRKibRYioo8=", + "lastModified": 1765964414, + "narHash": "sha256-hIbtnxfAHSWa5m71lPVB/LN/WW4fUYojqBLDJ836naw=", "owner": "huggingface", "repo": "kernel-builder", - "rev": "967c94ec67830c5d85dc981407478939edd169f9", + "rev": "472f3a7fd9b2a16cf9801ecf70210e11376710d2", "type": "github" }, "original": { "owner": "huggingface", + "ref": "redo-capabilities", "repo": "kernel-builder", "type": "github" } }, "nixpkgs": { "locked": { - "lastModified": 1755963616, - "narHash": "sha256-6yD0ww/S8n+U2uPYcJZ3DRURP8Kx036GRpR2uPNZroE=", - "owner": "nixos", + "lastModified": 1763291491, + "narHash": "sha256-eEYvm+45PPmy+Qe+nZDpn1uhoMUjJwx3PwVVQoO9ksA=", + "owner": "NixOS", "repo": "nixpkgs", - "rev": "73e96df7cff5783f45e21342a75a1540c4eddce4", + "rev": "c543a59edf25ada193719764f3bc0c6ba835f94d", "type": "github" }, "original": { - "owner": "nixos", - "ref": "nixos-unstable-small", + "owner": "NixOS", "repo": "nixpkgs", + "rev": "c543a59edf25ada193719764f3bc0c6ba835f94d", "type": "github" } }, @@ -146,21 +89,6 @@ "repo": "default", "type": "github" } - }, - "systems_2": { - "locked": { - "lastModified": 1681028828, - "narHash": "sha256-Vy1rq5AaRuLzOxct8nz4T6wlgyUR7zLU309k9mBC768=", - "owner": "nix-systems", - "repo": "default", - "rev": "da67096a3b9bf56a91d16901293e51ba5b49a27e", - "type": "github" - }, - "original": { - "owner": "nix-systems", - "repo": "default", - "type": "github" - } } }, "root": "root", diff --git a/flake.nix b/flake.nix index ee6d68ec2b4e0ae4c7a9139538f9706cb31b1d2e..525d45a5e717b4ff007d835af96c14cfffec28e8 100644 --- a/flake.nix +++ b/flake.nix @@ -2,7 +2,7 @@ description = "Flake for Torch kernel extension"; inputs = { - kernel-builder.url = "github:huggingface/kernel-builder"; + kernel-builder.url = "github:huggingface/kernel-builder/redo-capabilities"; }; outputs = { self, kernel-builder, }: diff --git a/nix-build.log b/nix-build.log deleted file mode 100644 index 94e123ed87c19e81cfc58cea52d3d18fd9089f35..0000000000000000000000000000000000000000 --- a/nix-build.log +++ /dev/null @@ -1,13519 +0,0 @@ -warning: Git tree '/home/ec2-user/dev/sage_attention' is dirty -warning: not writing modified lock file of flake 'git+file:///home/ec2-user/dev/sage_attention': -• Updated input 'kernel-builder': - 'github:huggingface/kernel-builder/967c94ec67830c5d85dc981407478939edd169f9?narHash=sha256-0EbrJkAx7yTOOjfJQFbk%2BBLo4MYfsD6JgRKibRYioo8%3D' (2025-09-25) - → 'github:huggingface/kernel-builder/9532ae833d245c03cb4daebd510e89e14cd27e7d?narHash=sha256-LYeNDsudfsy7extv59oyyirOv0%2BCG4hPIvTKnUaG7m0%3D' (2025-09-30) -evaluation warning: `rev` argument of `genFlakeOutputs` is deprecated, pass `self` as follows: - - kernel-builder.lib.genFlakeOutputs { - inherit self; - path = ./.; - }; -these 7 derivations will be built: - /nix/store/66w6nb3dq6bjb0bils6wjgx4jz6bhchj-sage_attention-torch-ext.drv - /nix/store/6i491c8acr1vm7lb8xfzddscbgzqyjwl-sage_attention-torch-ext.drv - /nix/store/hdl8in3dz7ijjx0l9sk7cd78a66z644f-sage_attention-torch-ext.drv - /nix/store/vrayd3xii2nak1sl95dn7cw4irmy21b5-sage_attention-torch-ext.drv - /nix/store/xqrfdlnc06wy3nl55ypzrqk470gldyn4-sage_attention-torch-ext.drv - /nix/store/ll29z68nd2l37pyv7ihc5sw3qv1wffjp-torch-ext-bundle.drv - /nix/store/dyhilzcxbx6rq6fhn6kpnhig1jf5rv4c-build-and-copy.drv -building '/nix/store/66w6nb3dq6bjb0bils6wjgx4jz6bhchj-sage_attention-torch-ext.drv'... -building '/nix/store/6i491c8acr1vm7lb8xfzddscbgzqyjwl-sage_attention-torch-ext.drv'... -building '/nix/store/hdl8in3dz7ijjx0l9sk7cd78a66z644f-sage_attention-torch-ext.drv'... -building '/nix/store/vrayd3xii2nak1sl95dn7cw4irmy21b5-sage_attention-torch-ext.drv'... -building '/nix/store/xqrfdlnc06wy3nl55ypzrqk470gldyn4-sage_attention-torch-ext.drv'... -sage_attention-torch-ext> Sourcing get-kernel-check-hook.sh -sage_attention-torch-ext> Sourcing setup-cuda-hook -sage_attention-torch-ext> Sourcing get-kernel-check-hook.sh -sage_attention-torch-ext> Sourcing setup-cuda-hook -sage_attention-torch-ext> Sourcing get-kernel-check-hook.sh -sage_attention-torch-ext> Sourcing setup-cuda-hook -sage_attention-torch-ext> Sourcing get-kernel-check-hook.sh -sage_attention-torch-ext> Sourcing setup-cuda-hook -sage_attention-torch-ext> Sourcing get-kernel-check-hook.sh -sage_attention-torch-ext> Sourcing setup-cuda-hook -sage_attention-torch-ext> Running phase: unpackPhase -sage_attention-torch-ext> unpacking source archive /nix/store/6d6az24clsfzyy97rmzcdk0qax7l7c4p-source -sage_attention-torch-ext> source root is source -sage_attention-torch-ext> Running phase: patchPhase -sage_attention-torch-ext> Running phase: updateAutotoolsGnuConfigScriptsPhase -sage_attention-torch-ext> Running phase: configurePhase -sage_attention-torch-ext> Running phase: unpackPhase -sage_attention-torch-ext> Running phase: unpackPhase -sage_attention-torch-ext> Executing setupCUDAToolkitCompilers -sage_attention-torch-ext> fixing cmake files... -sage_attention-torch-ext> unpacking source archive /nix/store/6d6az24clsfzyy97rmzcdk0qax7l7c4p-source -sage_attention-torch-ext> unpacking source archive /nix/store/6d6az24clsfzyy97rmzcdk0qax7l7c4p-source -sage_attention-torch-ext> source root is source -sage_attention-torch-ext> source root is source -sage_attention-torch-ext> Running phase: unpackPhase -sage_attention-torch-ext> Running phase: patchPhase -sage_attention-torch-ext> Running phase: patchPhase -sage_attention-torch-ext> unpacking source archive /nix/store/6d6az24clsfzyy97rmzcdk0qax7l7c4p-source -sage_attention-torch-ext> source root is source -sage_attention-torch-ext> Running phase: unpackPhase -sage_attention-torch-ext> Running phase: updateAutotoolsGnuConfigScriptsPhase -sage_attention-torch-ext> Running phase: updateAutotoolsGnuConfigScriptsPhase -sage_attention-torch-ext> unpacking source archive /nix/store/6d6az24clsfzyy97rmzcdk0qax7l7c4p-source -sage_attention-torch-ext> Running phase: patchPhase -sage_attention-torch-ext> Running phase: configurePhase -sage_attention-torch-ext> cmake flags: -GNinja -DCMAKE_FIND_USE_SYSTEM_PACKAGE_REGISTRY=OFF -DCMAKE_FIND_USE_PACKAGE_REGISTRY=OFF -DCMAKE_EXPORT_NO_PACKAGE_REGISTRY=ON -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=OFF -DCMAKE_INSTALL_LOCALEDIR=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/share/locale -DCMAKE_INSTALL_LIBEXECDIR=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/libexec -DCMAKE_INSTALL_LIBDIR=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/lib -DCMAKE_INSTALL_DOCDIR=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/share/doc/sage_attention -DCMAKE_INSTALL_INFODIR=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/share/info -DCMAKE_INSTALL_MANDIR=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/share/man -DCMAKE_INSTALL_INCLUDEDIR=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/include -DCMAKE_INSTALL_SBINDIR=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/sbin -DCMAKE_INSTALL_BINDIR=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/bin -DCMAKE_INSTALL_NAME_DIR=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/lib -DCMAKE_POLICY_DEFAULT_CMP0025=NEW -DCMAKE_FIND_FRAMEWORK=LAST -DCMAKE_STRIP=/nix/store/rgfv9lch0b6ksjzlzsx0mljsb0ypqr8x-gcc-wrapper-13.4.0/bin/strip -DCMAKE_RANLIB=/nix/store/rgfv9lch0b6ksjzlzsx0mljsb0ypqr8x-gcc-wrapper-13.4.0/bin/ranlib -DCMAKE_AR=/nix/store/rgfv9lch0b6ksjzlzsx0mljsb0ypqr8x-gcc-wrapper-13.4.0/bin/ar -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_INSTALL_PREFIX=/nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext -DPython_EXECUTABLE:STRING=/nix/store/r3gwdvvsgl1csl12f4pkhz0jhsch7bdy-python3-3.13.6-env/bin/python -DCMAKE_CUDA_HOST_COMPILER:STRING=/nix/store/rgfv9lch0b6ksjzlzsx0mljsb0ypqr8x-gcc-wrapper-13.4.0/bin/g++ -DNVCC_THREADS=3 -DCUDAToolkit_INCLUDE_DIR=/nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85/include\;/nix/store/5f6h6xs5c74iqcjda3y73i290mfwfs9x-cuda_nvml_dev-12.6.77-dev/include\;/nix/store/r26q9f2lhsvimxha44g1xcck3adrdqwg-cuda_nvrtc-12.6.85-dev/include\;/nix/store/9ik1skjb698l6vkx4m4wvx2nrr4sx0na-libcufft-11.3.0.4-dev/include\;/nix/store/vl1dficb0blxzqg6xqzfi5p119jvl2vi-libcusolver-11.7.1.2-dev/include\;/nix/store/n7x9kkzi2jdfj6f6yjwywfhyfmn957zp-cuda_cupti-12.6.80-dev/include\;/nix/store/sskxmb670akk0avrahrl4r6hp7925zh8-cuda_cudart-12.6.77-dev/include\;/nix/store/8a9vz66yzsar01lpgipmzq8skyk3ymkp-cuda_cccl-12.6.77-dev/include\;/nix/store/xd2xrldv3lbg1bk93nr0yccy6j0vhh2k-cudnn-9.11.0.98-dev/include\;/nix/store/0w4g3rxgkw9r0lv737rslqdk7wldmi0n-libcurand-10.3.7.77-dev/include\;/nix/store/m0s4p867fk6wk8ba7ym9yff4mayqjhlw-libcusparse-12.5.4.2-dev/include\;/nix/store/blh9iyvjkmwd871mfjvfhnp7njwgnc6b-cuda_profiler_api-12.6.77-dev/include\;/nix/store/fy71fffqbwg3xgvygn66kd4igj65gblv-libcublas-12.6.4.1-dev/include\;/nix/store/4pwy3k2s52ppzbs3k6d58kda8jhmiim4-libcufile-1.11.1.6-dev/include -DCUDAToolkit_ROOT=/nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85\;/nix/store/1qgrl2sgdj5m7llm2vs9690gd9998psq-cudnn-9.11.0.98\;/nix/store/d2z15dzsgfm4r2yyl16n3wc0sw8z6fia-cuda_cupti-12.6.80-lib\;/nix/store/86ngm5djfbl6a0i43j282680chqz1vr8-libcusparse-12.5.4.2-lib\;/nix/store/bmph9rbyqnyjs02zriwq78kg16h12wi6-libcublas-12.6.4.1-lib\;/nix/store/wny8xmyma0ziffas96ansxgmjfqpw393-cuda_nvrtc-12.6.85-lib\;/nix/store/j40ndiqjiqbiqrbfmgmkzz6w8757cgvk-cuda_nvml_dev-12.6.77-lib\;/nix/store/3ii532blh586xxavim32i21kr84wlcdc-cuda_profiler_api-12.6.77\;/nix/store/j32l8jnzckhdy2lzxgyd59y7p39y6b1d-libcusolver-11.7.1.2-static\;/nix/store/5iv2zpbf4k00ch4c5zfi5b8dlj90y3d3-cuda_cccl-12.6.77\;/nix/store/a8yi28jqv5185bbv10jpjja3x98i86hm-cuda_cudart-12.6.77-stubs\;/nix/store/ya85qn68jv6mlq6gh6phh5hwk3dkynag-cuda_cudart-12.6.77-static\;/nix/store/m65ribrsnk3gbabcx9ah6phgiil19j01-libcufile-1.11.1.6\;/nix/store/5f6h6xs5c74iqcjda3y73i290mfwfs9x-cuda_nvml_dev-12.6.77-dev\;/nix/store/r26q9f2lhsvimxha44g1xcck3adrdqwg-cuda_nvrtc-12.6.85-dev\;/nix/store/9ik1skjb698l6vkx4m4wvx2nrr4sx0na-libcufft-11.3.0.4-dev\;/nix/store/k5rbpivsz3ilsxg91pgigp6la8ln3cv9-cuda_cupti-12.6.80\;/nix/store/vl1dficb0blxzqg6xqzfi5p119jvl2vi-libcusolver-11.7.1.2-dev\;/nix/store/n7x9kkzi2jdfj6f6yjwywfhyfmn957zp-cuda_cupti-12.6.80-dev\;/nix/store/m0fwdgh4nmrjd0q9v4m2ly63qbcq2hi2-cuda_cudart-12.6.77\;/nix/store/qfaxx4b8l1alrrl0gbyb23k3j850c0v5-libcurand-10.3.7.77-static\;/nix/store/w1npzy8mfl28w7cib5idkg6nvlbzhpzq-libcufile-1.11.1.6-lib\;/nix/store/8abbm2gd77dv0l3acw0s18wln36aa0l5-cuda_cudart-12.6.77-lib\;/nix/store/ykb9bv2lqkf1wzy73q96cb04pybx9xa2-cuda_nvcc-12.6.85-static\;/nix/store/nw9ws2qvhgdb33qgfx4iqj517814qq8y-libcufft-11.3.0.4\;/nix/store/sskxmb670akk0avrahrl4r6hp7925zh8-cuda_cudart-12.6.77-dev\;/nix/store/mfc3ah6lwfd8dfbs77b0z9i75c471b0n-libcufft-11.3.0.4-static\;/nix/store/zk3cg1ws6cskrzyhdr5d68f8zrkfk77d-cuda_nvrtc-12.6.85-static\;/nix/store/pcrirrvn2ya5d3r1y18s2zj4pm2jladw-libcusolver-11.7.1.2\;/nix/store/qdn67x8jrwr418air16kwicya4d747pq-libcufft-11.3.0.4-lib\;/nix/store/dg8hyrzy7sh3wdhcr4ywsz05cvl6vfyc-libcusparse-12.5.4.2\;/nix/store/8a9vz66yzsar01lpgipmzq8skyk3ymkp-cuda_cccl-12.6.77-dev\;/nix/store/wmcrrdxd3db58nklyp7yf90kknfdx6b5-libcurand-10.3.7.77-lib\;/nix/store/xd2xrldv3lbg1bk93nr0yccy6j0vhh2k-cudnn-9.11.0.98-dev\;/nix/store/0w4g3rxgkw9r0lv737rslqdk7wldmi0n-libcurand-10.3.7.77-dev\;/nix/store/jr1397g6pshvil5n4lnvp7dm24dm71h8-libcublas-12.6.4.1-static\;/nix/store/wq0wv7df58h6bgggnz964sk8m1hbkxxp-cuda_cupti-12.6.80-sample\;/nix/store/m0s4p867fk6wk8ba7ym9yff4mayqjhlw-libcusparse-12.5.4.2-dev\;/nix/store/blh9iyvjkmwd871mfjvfhnp7njwgnc6b-cuda_profiler_api-12.6.77-dev\;/nix/store/ngwsphsxf906z7cgwg32d1w83p809ywl-cudnn-9.11.0.98-static\;/nix/store/07zlxn68jyf4s263xafnjid55grmi7a2-cuda_nvrtc-12.6.85\;/nix/store/zyh7hqq402zc7dhafhbh9vycyzcfq256-libcurand-10.3.7.77\;/nix/store/x7mww4k0zzzb7bnffv0b22jqbyf1mg3v-cuda_cupti-12.6.80-static\;/nix/store/xvlapjc6spss1kvbjlq97m6pk19hfrxz-cuda_nvml_dev-12.6.77\;/nix/store/7j4zf0r8flh7l4x5pm1mgqb2vcabmcdj-libcusolver-11.7.1.2-lib\;/nix/store/gs8gw8bgjccrjxlyzhxa7h85gkxgqwhn-libcufile-1.11.1.6-static\;/nix/store/p9dnsv7mv8mqm9aisrckq8lm3zs3l7dk-cudnn-9.11.0.98-lib\;/nix/store/fy71fffqbwg3xgvygn66kd4igj65gblv-libcublas-12.6.4.1-dev\;/nix/store/dpska4iiya4xa5zzzmqzx3ljws73bnds-cuda_nvml_dev-12.6.77-static\;/nix/store/gzykkbwmch7pxgfzf86fg0b928lz6b36-libcusparse-12.5.4.2-static\;/nix/store/nqn7lvw8gbwbymdhz4nak9wf9b5bbah9-libcublas-12.6.4.1\;/nix/store/4pwy3k2s52ppzbs3k6d58kda8jhmiim4-libcufile-1.11.1.6-dev -DPROTOC_EXE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DProtobuf_PROTOC_EXE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DProtobuf_PROTOC_EXECUTABLE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DPYBIND11_PYTHONLIBS_OVERWRITE=OFF -DPYTHON_EXECUTABLE=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/bin/python3.13 -DPYTHON_INCLUDE_DIR=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/include/python3.13 -DPYTHON_SITE_PACKAGES=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/lib/python3.13/site-packages -sage_attention-torch-ext> Running phase: configurePhase -sage_attention-torch-ext> source root is source -sage_attention-torch-ext> Running phase: updateAutotoolsGnuConfigScriptsPhase -sage_attention-torch-ext> Running phase: patchPhase -sage_attention-torch-ext> Executing setupCUDAToolkitCompilers -sage_attention-torch-ext> fixing cmake files... -sage_attention-torch-ext> Executing setupCUDAToolkitCompilers -sage_attention-torch-ext> fixing cmake files... -sage_attention-torch-ext> Running phase: configurePhase -sage_attention-torch-ext> Running phase: updateAutotoolsGnuConfigScriptsPhase -sage_attention-torch-ext> Executing setupCUDAToolkitCompilers -sage_attention-torch-ext> Running phase: configurePhase -sage_attention-torch-ext> fixing cmake files... -sage_attention-torch-ext> Executing setupCUDAToolkitCompilers -sage_attention-torch-ext> cmake flags: -GNinja -DCMAKE_FIND_USE_SYSTEM_PACKAGE_REGISTRY=OFF -DCMAKE_FIND_USE_PACKAGE_REGISTRY=OFF -DCMAKE_EXPORT_NO_PACKAGE_REGISTRY=ON -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=OFF -DCMAKE_INSTALL_LOCALEDIR=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/share/locale -DCMAKE_INSTALL_LIBEXECDIR=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/libexec -DCMAKE_INSTALL_LIBDIR=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/lib -DCMAKE_INSTALL_DOCDIR=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/share/doc/sage_attention -DCMAKE_INSTALL_INFODIR=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/share/info -DCMAKE_INSTALL_MANDIR=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/share/man -DCMAKE_INSTALL_INCLUDEDIR=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/include -DCMAKE_INSTALL_SBINDIR=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/sbin -DCMAKE_INSTALL_BINDIR=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/bin -DCMAKE_INSTALL_NAME_DIR=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/lib -DCMAKE_POLICY_DEFAULT_CMP0025=NEW -DCMAKE_FIND_FRAMEWORK=LAST -DCMAKE_STRIP=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/strip -DCMAKE_RANLIB=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/ranlib -DCMAKE_AR=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/ar -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_INSTALL_PREFIX=/nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext -DPython_EXECUTABLE:STRING=/nix/store/qal2apcjwlw2p2kk05dwqdgzh8ml687l-python3-3.13.6-env/bin/python -DCMAKE_CUDA_HOST_COMPILER:STRING=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/g++ -DNVCC_THREADS=3 -DCUDAToolkit_INCLUDE_DIR=/nix/store/2dc9bgppqvyd6bd5m4j9zphiyhhd39lv-libcurand-10.3.9.90-dev/include\;/nix/store/x6d389mfn7v413ia2had715g7rdgghgm-cuda_nvrtc-12.8.93-dev/include\;/nix/store/4sz65s9xk80q9jij0i4zbp9xd1pmr3ja-libcusparse-12.5.8.93-dev/include\;/nix/store/11bshw90q985bpd9ds649qmgg0x54q7x-cudnn-9.11.0.98-dev/include\;/nix/store/8dwjdyr7y3dkqlgswpn9swz884lx62gf-cuda_cccl-12.8.90-dev/include\;/nix/store/4cq7zkla3djm6g5gkpzzx4gfikda2k7z-cuda_profiler_api-12.8.90-dev/include\;/nix/store/90nghg4zsrw6gki8y8hw4id3p31bc8rk-libcusolver-11.7.3.90-dev/include\;/nix/store/vg32acb8vlqyhkhabbgvmralfw0kwhi3-cuda_cudart-12.8.90-dev/include\;/nix/store/vqg4r8izl1fy2smmw4dwv4x1adkj0rfb-libcufft-11.3.3.83-dev/include\;/nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93/include\;/nix/store/5pvax5f2dg278j43b4llkdxim9y0bjaf-cuda_nvml_dev-12.8.90-dev/include\;/nix/store/mps4gsnyk6s676zadvcykjxn08yghk5a-libcufile-1.13.1.3-dev/include\;/nix/store/gz9xyhflw755r8fcxkc816fp54sj0hl4-cuda_cupti-12.8.90-dev/include\;/nix/store/qa4d2v0lsm6giyr4b4421qsdygz0yrrh-libcublas-12.8.4.1-dev/include -DCUDAToolkit_ROOT=/nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93\;/nix/store/w96jlfiy431jnsww1x3ak3chhssa3i2s-libcusparse-12.5.8.93\;/nix/store/6zj6v3b9v8xdjs94iq1228slqwr757ij-libcublas-12.8.4.1\;/nix/store/q85pndpvaqdznfijmkn0mlfp8y3v08dl-cuda_cccl-12.8.90\;/nix/store/2dc9bgppqvyd6bd5m4j9zphiyhhd39lv-libcurand-10.3.9.90-dev\;/nix/store/cwy7010iwla9b2v1fx82sp66v12r913x-libcublas-12.8.4.1-lib\;/nix/store/x6d389mfn7v413ia2had715g7rdgghgm-cuda_nvrtc-12.8.93-dev\;/nix/store/22n25ss46s0hgspdp26qk025w9m393cd-libcublas-12.8.4.1-static\;/nix/store/sc5wnfvmk0j73xdppxj25kgk8s98lscs-cuda_nvrtc-12.8.93-lib\;/nix/store/54wqrrh6qbrwmv2wkz6b216ljrqbhcji-cudnn-9.11.0.98\;/nix/store/4sz65s9xk80q9jij0i4zbp9xd1pmr3ja-libcusparse-12.5.8.93-dev\;/nix/store/11bshw90q985bpd9ds649qmgg0x54q7x-cudnn-9.11.0.98-dev\;/nix/store/8dwjdyr7y3dkqlgswpn9swz884lx62gf-cuda_cccl-12.8.90-dev\;/nix/store/1v8m3gdw08hnbs7qa4jbkflm9lg1r5q6-libcurand-10.3.9.90\;/nix/store/jc58pv1cxhvpblrnzgaai60x04q6m0bp-cuda_nvml_dev-12.8.90-lib\;/nix/store/khwhv5d4kmzjpsm785iz3sva6i9sj9r5-libcufile-1.13.1.3-static\;/nix/store/xv6c2jcc3adyqks2xl28p4r0q1g4bc92-cuda_cupti-12.8.90\;/nix/store/a2h2yfjfx0si8smnqmghw7ccj0qbnv81-cuda_cupti-12.8.90-lib\;/nix/store/4cq7zkla3djm6g5gkpzzx4gfikda2k7z-cuda_profiler_api-12.8.90-dev\;/nix/store/xccbzbpcn8r506zdvhvbkqkilhlrh3c5-cuda_cudart-12.8.90-lib\;/nix/store/acbir62i1d7kvka4plmxsq8442z7r1l2-cuda_cudart-12.8.90-stubs\;/nix/store/ckkcbggf4x93zg3xn9xr00jgxs2x5p21-cuda_nvml_dev-12.8.90-static\;/nix/store/ml3bkm8bz1lnjmfd8lyxbjqpi1llasr2-libcusolver-11.7.3.90\;/nix/store/9zlrjnq7lisarny3llszk131vy816x2w-libcufile-1.13.1.3\;/nix/store/90nghg4zsrw6gki8y8hw4id3p31bc8rk-libcusolver-11.7.3.90-dev\;/nix/store/vg32acb8vlqyhkhabbgvmralfw0kwhi3-cuda_cudart-12.8.90-dev\;/nix/store/y27d2s3rcw8d17wcw23glhlj5rhs8d6y-cuda_cudart-12.8.90\;/nix/store/n96pib9yj31n031dmrrx43m61js1r5rn-cuda_nvcc-12.8.93-static\;/nix/store/pabakly3280dnghh3i89wklfm61raf7z-cuda_cupti-12.8.90-sample\;/nix/store/l0jiwp1f0dhigd41qqf408c5qyabz2vd-cudnn-9.11.0.98-static\;/nix/store/95lzbxp68m127n6hyllbr3dh2mlj7y8m-libcufft-11.3.3.83\;/nix/store/lxsd5l6hnqcfgqc1nsn8mmmpx385m3k8-libcusparse-12.5.8.93-lib\;/nix/store/vqg4r8izl1fy2smmw4dwv4x1adkj0rfb-libcufft-11.3.3.83-dev\;/nix/store/4b9rdinnksj1856siw3qmwi9f10480ii-cuda_nvrtc-12.8.93-static\;/nix/store/qh7zggir1ikzh3kvkhi2mqzpyisl4153-libcurand-10.3.9.90-static\;/nix/store/n25l4gcpw8cry4rg2a4c9jw3f53i65zd-libcusolver-11.7.3.90-lib\;/nix/store/xh73kc8spwfvd6w6wc63pyq3zm6qlrja-cuda_nvml_dev-12.8.90\;/nix/store/bgiqy1z8588hgcdzyh9brhc015w3nii0-libcurand-10.3.9.90-lib\;/nix/store/5pvax5f2dg278j43b4llkdxim9y0bjaf-cuda_nvml_dev-12.8.90-dev\;/nix/store/7lf23alvk7yh64flf2mj6smx66sqyz9d-libcufile-1.13.1.3-lib\;/nix/store/lfqj2ni7r0ir3n840b8r1lh63mnqr0ar-libcusparse-12.5.8.93-static\;/nix/store/qmw5pq21avnfvsk657k0zr4nsgwxa4jm-cuda_cudart-12.8.90-static\;/nix/store/826d39r2b4gwafqsyhvzq2bmqv8ygzrd-cuda_profiler_api-12.8.90\;/nix/store/g52lygjflrsyr6wahpf0rvs3fpna3wq9-cudnn-9.11.0.98-lib\;/nix/store/gxw5c9f7q2f1pmy0g1zyblb8p2p891a4-libcufft-11.3.3.83-lib\;/nix/store/pbsi8w1in7q44z83ndqsaxyzfrr2frgh-cuda_nvrtc-12.8.93\;/nix/store/mps4gsnyk6s676zadvcykjxn08yghk5a-libcufile-1.13.1.3-dev\;/nix/store/mvfnbb1m20fkv2n0j69ky9s9afn8p7h1-libcufft-11.3.3.83-static\;/nix/store/8byjxgnvhcyav2283wcxp752d8280c36-libcusolver-11.7.3.90-static\;/nix/store/gz9xyhflw755r8fcxkc816fp54sj0hl4-cuda_cupti-12.8.90-dev\;/nix/store/jyd8jp3q1d408n8842rb8g6ziviwm7q1-cuda_cupti-12.8.90-static\;/nix/store/qa4d2v0lsm6giyr4b4421qsdygz0yrrh-libcublas-12.8.4.1-dev -DPROTOC_EXE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DProtobuf_PROTOC_EXE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DProtobuf_PROTOC_EXECUTABLE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DPYBIND11_PYTHONLIBS_OVERWRITE=OFF -DPYTHON_EXECUTABLE=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/bin/python3.13 -DPYTHON_INCLUDE_DIR=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/include/python3.13 -DPYTHON_SITE_PACKAGES=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/lib/python3.13/site-packages -sage_attention-torch-ext> fixing cmake files... -sage_attention-torch-ext> cmake flags: -GNinja -DCMAKE_FIND_USE_SYSTEM_PACKAGE_REGISTRY=OFF -DCMAKE_FIND_USE_PACKAGE_REGISTRY=OFF -DCMAKE_EXPORT_NO_PACKAGE_REGISTRY=ON -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=OFF -DCMAKE_INSTALL_LOCALEDIR=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/share/locale -DCMAKE_INSTALL_LIBEXECDIR=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/libexec -DCMAKE_INSTALL_LIBDIR=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/lib -DCMAKE_INSTALL_DOCDIR=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/share/doc/sage_attention -DCMAKE_INSTALL_INFODIR=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/share/info -DCMAKE_INSTALL_MANDIR=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/share/man -DCMAKE_INSTALL_INCLUDEDIR=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/include -DCMAKE_INSTALL_SBINDIR=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/sbin -DCMAKE_INSTALL_BINDIR=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/bin -DCMAKE_INSTALL_NAME_DIR=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/lib -DCMAKE_POLICY_DEFAULT_CMP0025=NEW -DCMAKE_FIND_FRAMEWORK=LAST -DCMAKE_STRIP=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/strip -DCMAKE_RANLIB=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/ranlib -DCMAKE_AR=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/ar -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_INSTALL_PREFIX=/nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext -DPython_EXECUTABLE:STRING=/nix/store/wirj6dihrpcch7idfd7jy4l0hqfsgkk1-python3-3.13.6-env/bin/python -DCMAKE_CUDA_HOST_COMPILER:STRING=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/g++ -DNVCC_THREADS=3 -DCUDAToolkit_INCLUDE_DIR=/nix/store/2dc9bgppqvyd6bd5m4j9zphiyhhd39lv-libcurand-10.3.9.90-dev/include\;/nix/store/x6d389mfn7v413ia2had715g7rdgghgm-cuda_nvrtc-12.8.93-dev/include\;/nix/store/4sz65s9xk80q9jij0i4zbp9xd1pmr3ja-libcusparse-12.5.8.93-dev/include\;/nix/store/11bshw90q985bpd9ds649qmgg0x54q7x-cudnn-9.11.0.98-dev/include\;/nix/store/8dwjdyr7y3dkqlgswpn9swz884lx62gf-cuda_cccl-12.8.90-dev/include\;/nix/store/4cq7zkla3djm6g5gkpzzx4gfikda2k7z-cuda_profiler_api-12.8.90-dev/include\;/nix/store/90nghg4zsrw6gki8y8hw4id3p31bc8rk-libcusolver-11.7.3.90-dev/include\;/nix/store/vg32acb8vlqyhkhabbgvmralfw0kwhi3-cuda_cudart-12.8.90-dev/include\;/nix/store/vqg4r8izl1fy2smmw4dwv4x1adkj0rfb-libcufft-11.3.3.83-dev/include\;/nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93/include\;/nix/store/5pvax5f2dg278j43b4llkdxim9y0bjaf-cuda_nvml_dev-12.8.90-dev/include\;/nix/store/klis291y7cza60yzgkxzbid80bnyshmr-cuda_nvtx-12.8.90-dev/include\;/nix/store/mps4gsnyk6s676zadvcykjxn08yghk5a-libcufile-1.13.1.3-dev/include\;/nix/store/gz9xyhflw755r8fcxkc816fp54sj0hl4-cuda_cupti-12.8.90-dev/include\;/nix/store/qa4d2v0lsm6giyr4b4421qsdygz0yrrh-libcublas-12.8.4.1-dev/include -DCUDAToolkit_ROOT=/nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93\;/nix/store/w96jlfiy431jnsww1x3ak3chhssa3i2s-libcusparse-12.5.8.93\;/nix/store/6zj6v3b9v8xdjs94iq1228slqwr757ij-libcublas-12.8.4.1\;/nix/store/q85pndpvaqdznfijmkn0mlfp8y3v08dl-cuda_cccl-12.8.90\;/nix/store/2dc9bgppqvyd6bd5m4j9zphiyhhd39lv-libcurand-10.3.9.90-dev\;/nix/store/cwy7010iwla9b2v1fx82sp66v12r913x-libcublas-12.8.4.1-lib\;/nix/store/x6d389mfn7v413ia2had715g7rdgghgm-cuda_nvrtc-12.8.93-dev\;/nix/store/22n25ss46s0hgspdp26qk025w9m393cd-libcublas-12.8.4.1-static\;/nix/store/sc5wnfvmk0j73xdppxj25kgk8s98lscs-cuda_nvrtc-12.8.93-lib\;/nix/store/54wqrrh6qbrwmv2wkz6b216ljrqbhcji-cudnn-9.11.0.98\;/nix/store/4sz65s9xk80q9jij0i4zbp9xd1pmr3ja-libcusparse-12.5.8.93-dev\;/nix/store/11bshw90q985bpd9ds649qmgg0x54q7x-cudnn-9.11.0.98-dev\;/nix/store/8dwjdyr7y3dkqlgswpn9swz884lx62gf-cuda_cccl-12.8.90-dev\;/nix/store/1v8m3gdw08hnbs7qa4jbkflm9lg1r5q6-libcurand-10.3.9.90\;/nix/store/jc58pv1cxhvpblrnzgaai60x04q6m0bp-cuda_nvml_dev-12.8.90-lib\;/nix/store/khwhv5d4kmzjpsm785iz3sva6i9sj9r5-libcufile-1.13.1.3-static\;/nix/store/xv6c2jcc3adyqks2xl28p4r0q1g4bc92-cuda_cupti-12.8.90\;/nix/store/a2h2yfjfx0si8smnqmghw7ccj0qbnv81-cuda_cupti-12.8.90-lib\;/nix/store/4cq7zkla3djm6g5gkpzzx4gfikda2k7z-cuda_profiler_api-12.8.90-dev\;/nix/store/5f6dvklv5d0mvygrrf0vzp0smcn7kk01-cuda_nvtx-12.8.90\;/nix/store/xccbzbpcn8r506zdvhvbkqkilhlrh3c5-cuda_cudart-12.8.90-lib\;/nix/store/acbir62i1d7kvka4plmxsq8442z7r1l2-cuda_cudart-12.8.90-stubs\;/nix/store/ckkcbggf4x93zg3xn9xr00jgxs2x5p21-cuda_nvml_dev-12.8.90-static\;/nix/store/ml3bkm8bz1lnjmfd8lyxbjqpi1llasr2-libcusolver-11.7.3.90\;/nix/store/9zlrjnq7lisarny3llszk131vy816x2w-libcufile-1.13.1.3\;/nix/store/90nghg4zsrw6gki8y8hw4id3p31bc8rk-libcusolver-11.7.3.90-dev\;/nix/store/vg32acb8vlqyhkhabbgvmralfw0kwhi3-cuda_cudart-12.8.90-dev\;/nix/store/y27d2s3rcw8d17wcw23glhlj5rhs8d6y-cuda_cudart-12.8.90\;/nix/store/wa9pr3485k3mw8jhv7i9kfzjrqmdl5bb-cuda_nvtx-12.8.90-lib\;/nix/store/n96pib9yj31n031dmrrx43m61js1r5rn-cuda_nvcc-12.8.93-static\;/nix/store/pabakly3280dnghh3i89wklfm61raf7z-cuda_cupti-12.8.90-sample\;/nix/store/l0jiwp1f0dhigd41qqf408c5qyabz2vd-cudnn-9.11.0.98-static\;/nix/store/95lzbxp68m127n6hyllbr3dh2mlj7y8m-libcufft-11.3.3.83\;/nix/store/lxsd5l6hnqcfgqc1nsn8mmmpx385m3k8-libcusparse-12.5.8.93-lib\;/nix/store/vqg4r8izl1fy2smmw4dwv4x1adkj0rfb-libcufft-11.3.3.83-dev\;/nix/store/4b9rdinnksj1856siw3qmwi9f10480ii-cuda_nvrtc-12.8.93-static\;/nix/store/qh7zggir1ikzh3kvkhi2mqzpyisl4153-libcurand-10.3.9.90-static\;/nix/store/n25l4gcpw8cry4rg2a4c9jw3f53i65zd-libcusolver-11.7.3.90-lib\;/nix/store/xh73kc8spwfvd6w6wc63pyq3zm6qlrja-cuda_nvml_dev-12.8.90\;/nix/store/bgiqy1z8588hgcdzyh9brhc015w3nii0-libcurand-10.3.9.90-lib\;/nix/store/5pvax5f2dg278j43b4llkdxim9y0bjaf-cuda_nvml_dev-12.8.90-dev\;/nix/store/7lf23alvk7yh64flf2mj6smx66sqyz9d-libcufile-1.13.1.3-lib\;/nix/store/klis291y7cza60yzgkxzbid80bnyshmr-cuda_nvtx-12.8.90-dev\;/nix/store/lfqj2ni7r0ir3n840b8r1lh63mnqr0ar-libcusparse-12.5.8.93-static\;/nix/store/qmw5pq21avnfvsk657k0zr4nsgwxa4jm-cuda_cudart-12.8.90-static\;/nix/store/826d39r2b4gwafqsyhvzq2bmqv8ygzrd-cuda_profiler_api-12.8.90\;/nix/store/g52lygjflrsyr6wahpf0rvs3fpna3wq9-cudnn-9.11.0.98-lib\;/nix/store/gxw5c9f7q2f1pmy0g1zyblb8p2p891a4-libcufft-11.3.3.83-lib\;/nix/store/pbsi8w1in7q44z83ndqsaxyzfrr2frgh-cuda_nvrtc-12.8.93\;/nix/store/mps4gsnyk6s676zadvcykjxn08yghk5a-libcufile-1.13.1.3-dev\;/nix/store/mvfnbb1m20fkv2n0j69ky9s9afn8p7h1-libcufft-11.3.3.83-static\;/nix/store/8byjxgnvhcyav2283wcxp752d8280c36-libcusolver-11.7.3.90-static\;/nix/store/gz9xyhflw755r8fcxkc816fp54sj0hl4-cuda_cupti-12.8.90-dev\;/nix/store/jyd8jp3q1d408n8842rb8g6ziviwm7q1-cuda_cupti-12.8.90-static\;/nix/store/qa4d2v0lsm6giyr4b4421qsdygz0yrrh-libcublas-12.8.4.1-dev -DPROTOC_EXE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DProtobuf_PROTOC_EXE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DProtobuf_PROTOC_EXECUTABLE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DPYBIND11_PYTHONLIBS_OVERWRITE=OFF -DPYTHON_EXECUTABLE=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/bin/python3.13 -DPYTHON_INCLUDE_DIR=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/include/python3.13 -DPYTHON_SITE_PACKAGES=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/lib/python3.13/site-packages -sage_attention-torch-ext> cmake flags: -GNinja -DCMAKE_FIND_USE_SYSTEM_PACKAGE_REGISTRY=OFF -DCMAKE_FIND_USE_PACKAGE_REGISTRY=OFF -DCMAKE_EXPORT_NO_PACKAGE_REGISTRY=ON -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=OFF -DCMAKE_INSTALL_LOCALEDIR=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/share/locale -DCMAKE_INSTALL_LIBEXECDIR=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/libexec -DCMAKE_INSTALL_LIBDIR=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/lib -DCMAKE_INSTALL_DOCDIR=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/share/doc/sage_attention -DCMAKE_INSTALL_INFODIR=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/share/info -DCMAKE_INSTALL_MANDIR=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/share/man -DCMAKE_INSTALL_INCLUDEDIR=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/include -DCMAKE_INSTALL_SBINDIR=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/sbin -DCMAKE_INSTALL_BINDIR=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/bin -DCMAKE_INSTALL_NAME_DIR=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/lib -DCMAKE_POLICY_DEFAULT_CMP0025=NEW -DCMAKE_FIND_FRAMEWORK=LAST -DCMAKE_STRIP=/nix/store/rgfv9lch0b6ksjzlzsx0mljsb0ypqr8x-gcc-wrapper-13.4.0/bin/strip -DCMAKE_RANLIB=/nix/store/rgfv9lch0b6ksjzlzsx0mljsb0ypqr8x-gcc-wrapper-13.4.0/bin/ranlib -DCMAKE_AR=/nix/store/rgfv9lch0b6ksjzlzsx0mljsb0ypqr8x-gcc-wrapper-13.4.0/bin/ar -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_INSTALL_PREFIX=/nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext -DPython_EXECUTABLE:STRING=/nix/store/j6r6hpjs8p5m4s3i8cqqavg62fd5z48g-python3-3.13.6-env/bin/python -DCMAKE_CUDA_HOST_COMPILER:STRING=/nix/store/rgfv9lch0b6ksjzlzsx0mljsb0ypqr8x-gcc-wrapper-13.4.0/bin/g++ -DNVCC_THREADS=3 -DCUDAToolkit_INCLUDE_DIR=/nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85/include\;/nix/store/5f6h6xs5c74iqcjda3y73i290mfwfs9x-cuda_nvml_dev-12.6.77-dev/include\;/nix/store/r26q9f2lhsvimxha44g1xcck3adrdqwg-cuda_nvrtc-12.6.85-dev/include\;/nix/store/nj1a061pvzpq9dr65yj3jpjqcx6pr4fq-cuda_nvtx-12.6.77-dev/include\;/nix/store/9ik1skjb698l6vkx4m4wvx2nrr4sx0na-libcufft-11.3.0.4-dev/include\;/nix/store/vl1dficb0blxzqg6xqzfi5p119jvl2vi-libcusolver-11.7.1.2-dev/include\;/nix/store/n7x9kkzi2jdfj6f6yjwywfhyfmn957zp-cuda_cupti-12.6.80-dev/include\;/nix/store/sskxmb670akk0avrahrl4r6hp7925zh8-cuda_cudart-12.6.77-dev/include\;/nix/store/8a9vz66yzsar01lpgipmzq8skyk3ymkp-cuda_cccl-12.6.77-dev/include\;/nix/store/xd2xrldv3lbg1bk93nr0yccy6j0vhh2k-cudnn-9.11.0.98-dev/include\;/nix/store/0w4g3rxgkw9r0lv737rslqdk7wldmi0n-libcurand-10.3.7.77-dev/include\;/nix/store/m0s4p867fk6wk8ba7ym9yff4mayqjhlw-libcusparse-12.5.4.2-dev/include\;/nix/store/blh9iyvjkmwd871mfjvfhnp7njwgnc6b-cuda_profiler_api-12.6.77-dev/include\;/nix/store/fy71fffqbwg3xgvygn66kd4igj65gblv-libcublas-12.6.4.1-dev/include\;/nix/store/4pwy3k2s52ppzbs3k6d58kda8jhmiim4-libcufile-1.11.1.6-dev/include -DCUDAToolkit_ROOT=/nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85\;/nix/store/1qgrl2sgdj5m7llm2vs9690gd9998psq-cudnn-9.11.0.98\;/nix/store/d2z15dzsgfm4r2yyl16n3wc0sw8z6fia-cuda_cupti-12.6.80-lib\;/nix/store/86ngm5djfbl6a0i43j282680chqz1vr8-libcusparse-12.5.4.2-lib\;/nix/store/bmph9rbyqnyjs02zriwq78kg16h12wi6-libcublas-12.6.4.1-lib\;/nix/store/wny8xmyma0ziffas96ansxgmjfqpw393-cuda_nvrtc-12.6.85-lib\;/nix/store/j40ndiqjiqbiqrbfmgmkzz6w8757cgvk-cuda_nvml_dev-12.6.77-lib\;/nix/store/3ii532blh586xxavim32i21kr84wlcdc-cuda_profiler_api-12.6.77\;/nix/store/j32l8jnzckhdy2lzxgyd59y7p39y6b1d-libcusolver-11.7.1.2-static\;/nix/store/5iv2zpbf4k00ch4c5zfi5b8dlj90y3d3-cuda_cccl-12.6.77\;/nix/store/a8yi28jqv5185bbv10jpjja3x98i86hm-cuda_cudart-12.6.77-stubs\;/nix/store/ya85qn68jv6mlq6gh6phh5hwk3dkynag-cuda_cudart-12.6.77-static\;/nix/store/m65ribrsnk3gbabcx9ah6phgiil19j01-libcufile-1.11.1.6\;/nix/store/5f6h6xs5c74iqcjda3y73i290mfwfs9x-cuda_nvml_dev-12.6.77-dev\;/nix/store/r26q9f2lhsvimxha44g1xcck3adrdqwg-cuda_nvrtc-12.6.85-dev\;/nix/store/nj1a061pvzpq9dr65yj3jpjqcx6pr4fq-cuda_nvtx-12.6.77-dev\;/nix/store/bcvj4g3f3n6cpb6czcb5k8zdmyd94fwi-cuda_nvtx-12.6.77-lib\;/nix/store/9ik1skjb698l6vkx4m4wvx2nrr4sx0na-libcufft-11.3.0.4-dev\;/nix/store/k5rbpivsz3ilsxg91pgigp6la8ln3cv9-cuda_cupti-12.6.80\;/nix/store/vl1dficb0blxzqg6xqzfi5p119jvl2vi-libcusolver-11.7.1.2-dev\;/nix/store/f87x0n0gi2d7rxh1ja92za2ixcw60q2p-cuda_nvtx-12.6.77\;/nix/store/n7x9kkzi2jdfj6f6yjwywfhyfmn957zp-cuda_cupti-12.6.80-dev\;/nix/store/m0fwdgh4nmrjd0q9v4m2ly63qbcq2hi2-cuda_cudart-12.6.77\;/nix/store/qfaxx4b8l1alrrl0gbyb23k3j850c0v5-libcurand-10.3.7.77-static\;/nix/store/w1npzy8mfl28w7cib5idkg6nvlbzhpzq-libcufile-1.11.1.6-lib\;/nix/store/8abbm2gd77dv0l3acw0s18wln36aa0l5-cuda_cudart-12.6.77-lib\;/nix/store/ykb9bv2lqkf1wzy73q96cb04pybx9xa2-cuda_nvcc-12.6.85-static\;/nix/store/nw9ws2qvhgdb33qgfx4iqj517814qq8y-libcufft-11.3.0.4\;/nix/store/sskxmb670akk0avrahrl4r6hp7925zh8-cuda_cudart-12.6.77-dev\;/nix/store/mfc3ah6lwfd8dfbs77b0z9i75c471b0n-libcufft-11.3.0.4-static\;/nix/store/zk3cg1ws6cskrzyhdr5d68f8zrkfk77d-cuda_nvrtc-12.6.85-static\;/nix/store/pcrirrvn2ya5d3r1y18s2zj4pm2jladw-libcusolver-11.7.1.2\;/nix/store/qdn67x8jrwr418air16kwicya4d747pq-libcufft-11.3.0.4-lib\;/nix/store/dg8hyrzy7sh3wdhcr4ywsz05cvl6vfyc-libcusparse-12.5.4.2\;/nix/store/8a9vz66yzsar01lpgipmzq8skyk3ymkp-cuda_cccl-12.6.77-dev\;/nix/store/wmcrrdxd3db58nklyp7yf90kknfdx6b5-libcurand-10.3.7.77-lib\;/nix/store/xd2xrldv3lbg1bk93nr0yccy6j0vhh2k-cudnn-9.11.0.98-dev\;/nix/store/0w4g3rxgkw9r0lv737rslqdk7wldmi0n-libcurand-10.3.7.77-dev\;/nix/store/jr1397g6pshvil5n4lnvp7dm24dm71h8-libcublas-12.6.4.1-static\;/nix/store/wq0wv7df58h6bgggnz964sk8m1hbkxxp-cuda_cupti-12.6.80-sample\;/nix/store/m0s4p867fk6wk8ba7ym9yff4mayqjhlw-libcusparse-12.5.4.2-dev\;/nix/store/blh9iyvjkmwd871mfjvfhnp7njwgnc6b-cuda_profiler_api-12.6.77-dev\;/nix/store/ngwsphsxf906z7cgwg32d1w83p809ywl-cudnn-9.11.0.98-static\;/nix/store/07zlxn68jyf4s263xafnjid55grmi7a2-cuda_nvrtc-12.6.85\;/nix/store/zyh7hqq402zc7dhafhbh9vycyzcfq256-libcurand-10.3.7.77\;/nix/store/x7mww4k0zzzb7bnffv0b22jqbyf1mg3v-cuda_cupti-12.6.80-static\;/nix/store/xvlapjc6spss1kvbjlq97m6pk19hfrxz-cuda_nvml_dev-12.6.77\;/nix/store/7j4zf0r8flh7l4x5pm1mgqb2vcabmcdj-libcusolver-11.7.1.2-lib\;/nix/store/gs8gw8bgjccrjxlyzhxa7h85gkxgqwhn-libcufile-1.11.1.6-static\;/nix/store/p9dnsv7mv8mqm9aisrckq8lm3zs3l7dk-cudnn-9.11.0.98-lib\;/nix/store/fy71fffqbwg3xgvygn66kd4igj65gblv-libcublas-12.6.4.1-dev\;/nix/store/dpska4iiya4xa5zzzmqzx3ljws73bnds-cuda_nvml_dev-12.6.77-static\;/nix/store/gzykkbwmch7pxgfzf86fg0b928lz6b36-libcusparse-12.5.4.2-static\;/nix/store/nqn7lvw8gbwbymdhz4nak9wf9b5bbah9-libcublas-12.6.4.1\;/nix/store/4pwy3k2s52ppzbs3k6d58kda8jhmiim4-libcufile-1.11.1.6-dev -DPROTOC_EXE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DProtobuf_PROTOC_EXE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DProtobuf_PROTOC_EXECUTABLE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DPYBIND11_PYTHONLIBS_OVERWRITE=OFF -DPYTHON_EXECUTABLE=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/bin/python3.13 -DPYTHON_INCLUDE_DIR=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/include/python3.13 -DPYTHON_SITE_PACKAGES=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/lib/python3.13/site-packages -sage_attention-torch-ext> cmake flags: -GNinja -DCMAKE_FIND_USE_SYSTEM_PACKAGE_REGISTRY=OFF -DCMAKE_FIND_USE_PACKAGE_REGISTRY=OFF -DCMAKE_EXPORT_NO_PACKAGE_REGISTRY=ON -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=OFF -DCMAKE_INSTALL_LOCALEDIR=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/share/locale -DCMAKE_INSTALL_LIBEXECDIR=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/libexec -DCMAKE_INSTALL_LIBDIR=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/lib -DCMAKE_INSTALL_DOCDIR=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/share/doc/sage_attention -DCMAKE_INSTALL_INFODIR=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/share/info -DCMAKE_INSTALL_MANDIR=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/share/man -DCMAKE_INSTALL_INCLUDEDIR=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/include -DCMAKE_INSTALL_SBINDIR=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/sbin -DCMAKE_INSTALL_BINDIR=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/bin -DCMAKE_INSTALL_NAME_DIR=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/lib -DCMAKE_POLICY_DEFAULT_CMP0025=NEW -DCMAKE_FIND_FRAMEWORK=LAST -DCMAKE_STRIP=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/strip -DCMAKE_RANLIB=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/ranlib -DCMAKE_AR=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/ar -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++ -DCMAKE_INSTALL_PREFIX=/nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext -DPython_EXECUTABLE:STRING=/nix/store/aikr517kmcd8r2nrrj70jq71d7352qiq-python3-3.13.6-env/bin/python -DCMAKE_CUDA_HOST_COMPILER:STRING=/nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/g++ -DNVCC_THREADS=3 -DCUDAToolkit_INCLUDE_DIR=/nix/store/kky5wd8qwb0hx3jb3j9qc1bkwznw3z83-libcusparse-12.5.10.65-dev/include\;/nix/store/dd8wl3nnsigw2gj5bwaiswla97jpw1jz-libcublas-12.9.1.4-dev/include\;/nix/store/zsmc0yjbjrfbamm9ycrlz5yzi5hrbag1-libcurand-10.3.10.19-dev/include\;/nix/store/ip4lb9ximc445dbdkdvia4whx83g00g3-libcusolver-11.7.5.82-dev/include\;/nix/store/81xppf0rrqfasvg7wy4z891ab473nb9v-libcufile-1.14.1.1-dev/include\;/nix/store/nkvyh0qxbfj2wbm3r800xd6x1fhs1s4x-cuda_cccl-12.9.27-dev/include\;/nix/store/ik96pdimvw3bjj8wdr6laxycnn5lpwby-libcufft-11.4.1.4-dev/include\;/nix/store/f9r19xpj8qayy3b74gx3gbjrq0z1aq3b-cuda_nvml_dev-12.9.79-dev/include\;/nix/store/0kycn0pb0x46h16afxw2bjrm1gjq1355-cuda_profiler_api-12.9.79-dev/include\;/nix/store/z2xfln4d3r92hjjihlq5w6hvh5qhpcb4-cudnn-9.11.0.98-dev/include\;/nix/store/x4w41r4jyapqwdghvi6xrpd0mnim4x08-cuda_cudart-12.9.79-dev/include\;/nix/store/8zrv6h6f2cfz34pwq012n4cx2zrv5m1s-cuda_nvcc-12.9.86/include\;/nix/store/f21f8hghg4fiwa2ix29h1zy854p7q4v6-cuda_nvrtc-12.9.86-dev/include\;/nix/store/ns0brisbkgrjyfi16rlyjjgcym4jk6qv-cuda_cupti-12.9.79-dev/include -DCUDAToolkit_ROOT=/nix/store/8zrv6h6f2cfz34pwq012n4cx2zrv5m1s-cuda_nvcc-12.9.86\;/nix/store/q2al0drhrl0yxk97xbsjl8d0h25kmsq9-libcurand-10.3.10.19-lib\;/nix/store/ax1ssn45048qbmyy19basgv6q64y5jy0-cuda_cupti-12.9.79\;/nix/store/m09542l6q83flp3asv2r4j3wcbjqksvg-libcufile-1.14.1.1-static\;/nix/store/b3wbcra9cziq8bwf3yhmj2nn1mf5bqy2-cuda_cudart-12.9.79-lib\;/nix/store/j5kp5fg9mn6hhslk18wbmskc7v96l353-cuda_cupti-12.9.79-static\;/nix/store/kky5wd8qwb0hx3jb3j9qc1bkwznw3z83-libcusparse-12.5.10.65-dev\;/nix/store/dd8wl3nnsigw2gj5bwaiswla97jpw1jz-libcublas-12.9.1.4-dev\;/nix/store/zsmc0yjbjrfbamm9ycrlz5yzi5hrbag1-libcurand-10.3.10.19-dev\;/nix/store/3s79bz4ldkhlks6jf9a2jd4r34y6018b-libcurand-10.3.10.19\;/nix/store/v48xzq66pzmygxqkws17n9nvpa7lad9d-cuda_nvml_dev-12.9.79\;/nix/store/6via2axi1n31n685jii6dwaiqca8b2rc-cuda_nvcc-12.9.86-static\;/nix/store/v0hx9fqdlmz9kvjd9sqr2zc141ny10yn-cuda_profiler_api-12.9.79\;/nix/store/ip4lb9ximc445dbdkdvia4whx83g00g3-libcusolver-11.7.5.82-dev\;/nix/store/8cig7k11qv5g8x0j8n2mbdfzwrnf7cg2-cuda_cudart-12.9.79-stubs\;/nix/store/xg8pj5m74n2h3v8kgxbvmbpcl90rzmlx-cudnn-9.11.0.98-static\;/nix/store/v4b7mkhyq1akczzkcyynj7y9c61l9dc7-cuda_cudart-12.9.79-static\;/nix/store/hw2swakbrvi4innrymcw8i2m98p73br0-cuda_cupti-12.9.79-sample\;/nix/store/s1i2kadnni2m4skpzzqzfzc3bpmrxi7p-libcusparse-12.5.10.65-lib\;/nix/store/81xppf0rrqfasvg7wy4z891ab473nb9v-libcufile-1.14.1.1-dev\;/nix/store/0a83zdhkh2i9d97r4zqdn8fi8vn4wfk3-libcublas-12.9.1.4-static\;/nix/store/nkvyh0qxbfj2wbm3r800xd6x1fhs1s4x-cuda_cccl-12.9.27-dev\;/nix/store/jnhjz87sm9nbnb72n54jj2l99szrzpg2-libcusparse-12.5.10.65\;/nix/store/ik96pdimvw3bjj8wdr6laxycnn5lpwby-libcufft-11.4.1.4-dev\;/nix/store/d1m6c5i6y6ncjygpdmv1b4pmd91hvjr2-cuda_cupti-12.9.79-lib\;/nix/store/49p6af3v11dcxvq9andr6l8csa2sr4j4-cuda_nvrtc-12.9.86-static\;/nix/store/bfygrgghga26l7br5d5j3h6hd1s21rkn-cudnn-9.11.0.98\;/nix/store/a6an9chi5dvjsybrfrxql0bn76xswzpa-libcufft-11.4.1.4\;/nix/store/f9r19xpj8qayy3b74gx3gbjrq0z1aq3b-cuda_nvml_dev-12.9.79-dev\;/nix/store/7zy91byrxpnyzhjlwham2gqyir2x6f54-libcusolver-11.7.5.82-lib\;/nix/store/0kycn0pb0x46h16afxw2bjrm1gjq1355-cuda_profiler_api-12.9.79-dev\;/nix/store/cx0hyla7fkqqc5hh1gn4hkarjyjvbjhf-libcusparse-12.5.10.65-static\;/nix/store/3yi8kx62nklnyn77zn4z23hi03l9c7ff-libcusolver-11.7.5.82-static\;/nix/store/z2xfln4d3r92hjjihlq5w6hvh5qhpcb4-cudnn-9.11.0.98-dev\;/nix/store/86nq76ks8vlgjdsnh1hkskyfw7mm3plc-cuda_cccl-12.9.27\;/nix/store/01ywykdxfkvp64318anifgx7zaavz9ql-cuda_nvml_dev-12.9.79-lib\;/nix/store/qv2m9i0nby2p03xx37mkkm84dlqb9s84-cuda_cudart-12.9.79\;/nix/store/a09saq5rl5jxbgv9gqllx0080ypjk00x-libcufile-1.14.1.1-lib\;/nix/store/0l18n4dhavr0p4rk0nyqqjr8paacak13-libcufile-1.14.1.1\;/nix/store/r8ly0w88qv4gw3lhd784ha0ag221c23s-cuda_nvrtc-12.9.86-lib\;/nix/store/rngn6cls1blhilrw78xb3pjgwghibhzk-libcurand-10.3.10.19-static\;/nix/store/x4w41r4jyapqwdghvi6xrpd0mnim4x08-cuda_cudart-12.9.79-dev\;/nix/store/ikw7sqic4kknjkp50dr54khgs06q1hbv-cuda_nvml_dev-12.9.79-static\;/nix/store/bzdnjn29xj8a73wg16qrz0sswi9svp0x-libcublas-12.9.1.4\;/nix/store/62hqkwasnanq5i1j63z4clc0s4c61k1r-libcufft-11.4.1.4-static\;/nix/store/5sjldyn2vmm4ky24v1f9ggs0hps496q3-libcusolver-11.7.5.82\;/nix/store/9c924z3749bfm078bwq4ad12kjz46pjf-libcufft-11.4.1.4-lib\;/nix/store/f21f8hghg4fiwa2ix29h1zy854p7q4v6-cuda_nvrtc-12.9.86-dev\;/nix/store/c1kdvq8xqqkwzzazl99w20h4x9z0f9pc-libcublas-12.9.1.4-lib\;/nix/store/ns0brisbkgrjyfi16rlyjjgcym4jk6qv-cuda_cupti-12.9.79-dev\;/nix/store/h6kzw3gvlv4sa0apb4fflpjlirhj72ga-cudnn-9.11.0.98-lib\;/nix/store/f5gvpjis5y727lw6vzr2h1zkb3hm08k2-cuda_nvrtc-12.9.86 -DPROTOC_EXE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DProtobuf_PROTOC_EXE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DProtobuf_PROTOC_EXECUTABLE=/nix/store/g82m0ia59azh4a1bcrk0r15qck6hp8da-protobuf-31.1/bin/protoc -DPYBIND11_PYTHONLIBS_OVERWRITE=OFF -DPYTHON_EXECUTABLE=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/bin/python3.13 -DPYTHON_INCLUDE_DIR=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/include/python3.13 -DPYTHON_SITE_PACKAGES=/nix/store/iyff8129iampdw13nlfqalzhxy8y1hi9-python3-3.13.6/lib/python3.13/site-packages -sage_attention-torch-ext> -- The CXX compiler identification is GNU 13.4.0 -sage_attention-torch-ext> -- Detecting CXX compiler ABI info -sage_attention-torch-ext> -- The CXX compiler identification is GNU 14.3.0 -sage_attention-torch-ext> -- The CXX compiler identification is GNU 14.3.0 -sage_attention-torch-ext> -- Detecting CXX compiler ABI info -sage_attention-torch-ext> -- The CXX compiler identification is GNU 13.4.0 -sage_attention-torch-ext> -- Detecting CXX compiler ABI info -sage_attention-torch-ext> -- Detecting CXX compiler ABI info -sage_attention-torch-ext> -- The CXX compiler identification is GNU 14.3.0 -sage_attention-torch-ext> -- Detecting CXX compiler ABI info -sage_attention-torch-ext> -- Detecting CXX compiler ABI info - done -sage_attention-torch-ext> -- Check for working CXX compiler: /nix/store/rgfv9lch0b6ksjzlzsx0mljsb0ypqr8x-gcc-wrapper-13.4.0/bin/g++ - skipped -sage_attention-torch-ext> -- Detecting CXX compile features -sage_attention-torch-ext> -- Detecting CXX compile features - done -sage_attention-torch-ext> -- Detecting CXX compiler ABI info - done -sage_attention-torch-ext> -- Detecting CXX compiler ABI info - done -sage_attention-torch-ext> -- Detecting CXX compiler ABI info - done -sage_attention-torch-ext> -- Detecting CXX compiler ABI info - done -sage_attention-torch-ext> -- Check for working CXX compiler: /nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/g++ - skipped -sage_attention-torch-ext> -- Detecting CXX compile features -sage_attention-torch-ext> -- Detecting CXX compile features - done -sage_attention-torch-ext> -- Check for working CXX compiler: /nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/g++ - skipped -sage_attention-torch-ext> -- Detecting CXX compile features -sage_attention-torch-ext> -- Detecting CXX compile features - done -sage_attention-torch-ext> -- FetchContent base directory: /build/source/build/_deps -sage_attention-torch-ext> -- Check for working CXX compiler: /nix/store/rgfv9lch0b6ksjzlzsx0mljsb0ypqr8x-gcc-wrapper-13.4.0/bin/g++ - skipped -sage_attention-torch-ext> -- Detecting CXX compile features -sage_attention-torch-ext> -- Detecting CXX compile features - done -sage_attention-torch-ext> -- Check for working CXX compiler: /nix/store/d8likaw8xxdmh2qmmasbm88h74q6a2gr-gcc-wrapper-14.3.0/bin/g++ - skipped -sage_attention-torch-ext> -- Detecting CXX compile features -sage_attention-torch-ext> -- Detecting CXX compile features - done -sage_attention-torch-ext> -- FetchContent base directory: /build/source/build/_deps -sage_attention-torch-ext> -- FetchContent base directory: /build/source/build/_deps -sage_attention-torch-ext> -- FetchContent base directory: /build/source/build/_deps -sage_attention-torch-ext> -- FetchContent base directory: /build/source/build/_deps -sage_attention-torch-ext> -- Found Python: /nix/store/r3gwdvvsgl1csl12f4pkhz0jhsch7bdy-python3-3.13.6-env/bin/python (found version "3.13.6") found components: Development Development.SABIModule Interpreter Development.Module Development.Embed -sage_attention-torch-ext> -- Found Python: /nix/store/wirj6dihrpcch7idfd7jy4l0hqfsgkk1-python3-3.13.6-env/bin/python (found version "3.13.6") found components: Development Development.SABIModule Interpreter Development.Module Development.Embed -sage_attention-torch-ext> -- Found Python: /nix/store/qal2apcjwlw2p2kk05dwqdgzh8ml687l-python3-3.13.6-env/bin/python (found version "3.13.6") found components: Development Development.SABIModule Interpreter Development.Module Development.Embed -sage_attention-torch-ext> -- Found Python: /nix/store/j6r6hpjs8p5m4s3i8cqqavg62fd5z48g-python3-3.13.6-env/bin/python (found version "3.13.6") found components: Development Development.SABIModule Interpreter Development.Module Development.Embed -sage_attention-torch-ext> -- Found Python: /nix/store/aikr517kmcd8r2nrrj70jq71d7352qiq-python3-3.13.6-env/bin/python (found version "3.13.6") found components: Development Development.SABIModule Interpreter Development.Module Development.Embed -sage_attention-torch-ext> -- Found CUDA: /nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93 (found version "12.8") -sage_attention-torch-ext> -- Found CUDA: /nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85 (found version "12.6") -sage_attention-torch-ext> -- Found CUDA: /nix/store/8zrv6h6f2cfz34pwq012n4cx2zrv5m1s-cuda_nvcc-12.9.86 (found version "12.9") -sage_attention-torch-ext> -- Found CUDA: /nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85 (found version "12.6") -sage_attention-torch-ext> -- Found CUDA: /nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93 (found version "12.8") -sage_attention-torch-ext> -- The CUDA compiler identification is NVIDIA 12.6.85 with host compiler GNU 13.4.0 -sage_attention-torch-ext> -- Detecting CUDA compiler ABI info -sage_attention-torch-ext> -- The CUDA compiler identification is NVIDIA 12.8.93 with host compiler GNU 14.3.0 -sage_attention-torch-ext> -- Detecting CUDA compiler ABI info -sage_attention-torch-ext> -- The CUDA compiler identification is NVIDIA 12.6.85 with host compiler GNU 13.4.0 -sage_attention-torch-ext> -- Detecting CUDA compiler ABI info -sage_attention-torch-ext> -- The CUDA compiler identification is NVIDIA 12.9.86 with host compiler GNU 14.3.0 -sage_attention-torch-ext> -- Detecting CUDA compiler ABI info -sage_attention-torch-ext> -- The CUDA compiler identification is NVIDIA 12.8.93 with host compiler GNU 14.3.0 -sage_attention-torch-ext> -- Detecting CUDA compiler ABI info -sage_attention-torch-ext> -- Detecting CUDA compiler ABI info - done -sage_attention-torch-ext> -- Check for working CUDA compiler: /nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85/bin/nvcc - skipped -sage_attention-torch-ext> -- Detecting CUDA compile features -sage_attention-torch-ext> -- Detecting CUDA compile features - done -sage_attention-torch-ext> -- Detecting CUDA compiler ABI info - done -sage_attention-torch-ext> -- Detecting CUDA compiler ABI info - done -sage_attention-torch-ext> -- Found CUDAToolkit: /nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85/include;/nix/store/fy71fffqbwg3xgvygn66kd4igj65gblv-libcublas-12.6.4.1-dev/include (found version "12.6.85") -sage_attention-torch-ext> -- Performing Test CMAKE_HAVE_LIBC_PTHREAD -sage_attention-torch-ext> -- Detecting CUDA compiler ABI info - done -sage_attention-torch-ext> -- Check for working CUDA compiler: /nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85/bin/nvcc - skipped -sage_attention-torch-ext> -- Check for working CUDA compiler: /nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93/bin/nvcc - skipped -sage_attention-torch-ext> -- Detecting CUDA compile features -sage_attention-torch-ext> -- Detecting CUDA compile features - done -sage_attention-torch-ext> -- Detecting CUDA compile features -sage_attention-torch-ext> -- Detecting CUDA compile features - done -sage_attention-torch-ext> -- Detecting CUDA compiler ABI info - done -sage_attention-torch-ext> -- Found CUDAToolkit: /nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85/include;/nix/store/fy71fffqbwg3xgvygn66kd4igj65gblv-libcublas-12.6.4.1-dev/include (found version "12.6.85") -sage_attention-torch-ext> -- Performing Test CMAKE_HAVE_LIBC_PTHREAD -sage_attention-torch-ext> -- Found CUDAToolkit: /nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93/include;/nix/store/qa4d2v0lsm6giyr4b4421qsdygz0yrrh-libcublas-12.8.4.1-dev/include (found version "12.8.93") -sage_attention-torch-ext> -- Performing Test CMAKE_HAVE_LIBC_PTHREAD -sage_attention-torch-ext> -- Check for working CUDA compiler: /nix/store/8zrv6h6f2cfz34pwq012n4cx2zrv5m1s-cuda_nvcc-12.9.86/bin/nvcc - skipped -sage_attention-torch-ext> -- Detecting CUDA compile features -sage_attention-torch-ext> -- Detecting CUDA compile features - done -sage_attention-torch-ext> -- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Failed -sage_attention-torch-ext> -- Looking for pthread_create in pthreads -sage_attention-torch-ext> -- Check for working CUDA compiler: /nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93/bin/nvcc - skipped -sage_attention-torch-ext> -- Detecting CUDA compile features -sage_attention-torch-ext> -- Detecting CUDA compile features - done -sage_attention-torch-ext> -- Found CUDAToolkit: /nix/store/8zrv6h6f2cfz34pwq012n4cx2zrv5m1s-cuda_nvcc-12.9.86/include;/nix/store/dd8wl3nnsigw2gj5bwaiswla97jpw1jz-libcublas-12.9.1.4-dev/include (found version "12.9.86") -sage_attention-torch-ext> -- Performing Test CMAKE_HAVE_LIBC_PTHREAD -sage_attention-torch-ext> -- Found CUDAToolkit: /nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93/include;/nix/store/qa4d2v0lsm6giyr4b4421qsdygz0yrrh-libcublas-12.8.4.1-dev/include (found version "12.8.93") -sage_attention-torch-ext> -- Performing Test CMAKE_HAVE_LIBC_PTHREAD -sage_attention-torch-ext> -- Looking for pthread_create in pthreads - not found -sage_attention-torch-ext> -- Looking for pthread_create in pthread -sage_attention-torch-ext> -- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Failed -sage_attention-torch-ext> -- Looking for pthread_create in pthreads -sage_attention-torch-ext> -- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Failed -sage_attention-torch-ext> -- Looking for pthread_create in pthreads -sage_attention-torch-ext> -- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Failed -sage_attention-torch-ext> -- Looking for pthread_create in pthreads -sage_attention-torch-ext> -- Looking for pthread_create in pthreads - not found -sage_attention-torch-ext> -- Looking for pthread_create in pthread -sage_attention-torch-ext> -- Looking for pthread_create in pthreads - not found -sage_attention-torch-ext> -- Looking for pthread_create in pthread -sage_attention-torch-ext> -- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Failed -sage_attention-torch-ext> -- Looking for pthread_create in pthreads -sage_attention-torch-ext> -- Looking for pthread_create in pthread - found -sage_attention-torch-ext> -- Found Threads: TRUE -sage_attention-torch-ext> -- Looking for pthread_create in pthreads - not found -sage_attention-torch-ext> -- Looking for pthread_create in pthread -sage_attention-torch-ext> -- Looking for pthread_create in pthreads - not found -sage_attention-torch-ext> -- Looking for pthread_create in pthread -sage_attention-torch-ext> -- Looking for pthread_create in pthread - found -sage_attention-torch-ext> -- Found Threads: TRUE -sage_attention-torch-ext> -- Looking for pthread_create in pthread - found -sage_attention-torch-ext> -- Found Threads: TRUE -sage_attention-torch-ext> -- Looking for pthread_create in pthread - found -sage_attention-torch-ext> -- Found Threads: TRUE -sage_attention-torch-ext> -- Looking for pthread_create in pthread - found -sage_attention-torch-ext> -- Found Threads: TRUE -sage_attention-torch-ext> -- PyTorch: CUDA detected: 12.6 -sage_attention-torch-ext> -- PyTorch: CUDA nvcc is: /nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85/bin/nvcc -sage_attention-torch-ext> -- PyTorch: CUDA toolkit directory: /nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85 -sage_attention-torch-ext> -- PyTorch: CUDA detected: 12.8 -sage_attention-torch-ext> -- PyTorch: CUDA nvcc is: /nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93/bin/nvcc -sage_attention-torch-ext> -- PyTorch: CUDA toolkit directory: /nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93 -sage_attention-torch-ext> -- PyTorch: CUDA detected: 12.6 -sage_attention-torch-ext> -- PyTorch: CUDA nvcc is: /nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85/bin/nvcc -sage_attention-torch-ext> -- PyTorch: CUDA toolkit directory: /nix/store/7iw4ipbdy17yzvqjhxpw03i17kq7f7rj-cuda_nvcc-12.6.85 -sage_attention-torch-ext> -- PyTorch: CUDA detected: 12.9 -sage_attention-torch-ext> -- PyTorch: CUDA nvcc is: /nix/store/8zrv6h6f2cfz34pwq012n4cx2zrv5m1s-cuda_nvcc-12.9.86/bin/nvcc -sage_attention-torch-ext> -- PyTorch: CUDA toolkit directory: /nix/store/8zrv6h6f2cfz34pwq012n4cx2zrv5m1s-cuda_nvcc-12.9.86 -sage_attention-torch-ext> -- PyTorch: Header version is: 12.6 -sage_attention-torch-ext> -- Found Python: /nix/store/r3gwdvvsgl1csl12f4pkhz0jhsch7bdy-python3-3.13.6-env/bin/python (found version "3.13.6") found components: Interpreter -sage_attention-torch-ext> CMake Warning at /nix/store/ld6fk094jhhsnbip1406vrky9lmyxbax-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:146 (message): -sage_attention-torch-ext> Failed to compute shorthash for libnvrtc.so -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/ld6fk094jhhsnbip1406vrky9lmyxbax-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/ld6fk094jhhsnbip1406vrky9lmyxbax-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- USE_CUDNN is set to 0. Compiling without cuDNN support -sage_attention-torch-ext> -- USE_CUSPARSELT is set to 0. Compiling without cuSPARSELt support -sage_attention-torch-ext> -- USE_CUDSS is set to 0. Compiling without cuDSS support -sage_attention-torch-ext> -- USE_CUFILE is set to 0. Compiling without cuFile support -sage_attention-torch-ext> -- Added CUDA NVCC flags for: -gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_89,code=sm_89;-gencode;arch=compute_90,code=sm_90 -sage_attention-torch-ext> CMake Warning at /nix/store/ld6fk094jhhsnbip1406vrky9lmyxbax-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:22 (message): -sage_attention-torch-ext> static library kineto_LIBRARY-NOTFOUND not found. -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/ld6fk094jhhsnbip1406vrky9lmyxbax-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:125 (append_torchlib_if_found) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- PyTorch: CUDA detected: 12.8 -sage_attention-torch-ext> -- PyTorch: CUDA nvcc is: /nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93/bin/nvcc -sage_attention-torch-ext> -- PyTorch: CUDA toolkit directory: /nix/store/8kyv8ffbfvksnqmm1kaz0llysg7dpn9z-cuda_nvcc-12.8.93 -sage_attention-torch-ext> -- Found Torch: /nix/store/pg32mpjmckfs38anjzgyvk2ljfw12pb3-python3.13-torch-2.8.0-lib/lib/libtorch.so -sage_attention-torch-ext> -- CUDA target architectures: 7.0;7.5;8.0;8.6;8.9;9.0 -sage_attention-torch-ext> -- CUDA supported target architectures: 7.0;7.5;8.0;8.6;8.9;9.0 -sage_attention-torch-ext> -- PyTorch: Header version is: 12.8 -sage_attention-torch-ext> -- PyTorch: Header version is: 12.6 -sage_attention-torch-ext> -- Found Python: /nix/store/qal2apcjwlw2p2kk05dwqdgzh8ml687l-python3-3.13.6-env/bin/python (found version "3.13.6") found components: Interpreter -sage_attention-torch-ext> -- Found Python: /nix/store/j6r6hpjs8p5m4s3i8cqqavg62fd5z48g-python3-3.13.6-env/bin/python (found version "3.13.6") found components: Interpreter -sage_attention-torch-ext> CMake Warning at /nix/store/6drs80sxjhskdki55g5k1dw0jzbd258w-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:146 (message): -sage_attention-torch-ext> Failed to compute shorthash for libnvrtc.so -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/6drs80sxjhskdki55g5k1dw0jzbd258w-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/6drs80sxjhskdki55g5k1dw0jzbd258w-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- USE_CUDNN is set to 0. Compiling without cuDNN support -sage_attention-torch-ext> -- USE_CUSPARSELT is set to 0. Compiling without cuSPARSELt support -sage_attention-torch-ext> -- USE_CUDSS is set to 0. Compiling without cuDSS support -sage_attention-torch-ext> -- USE_CUFILE is set to 0. Compiling without cuFile support -sage_attention-torch-ext> -- Added CUDA NVCC flags for: -gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_89,code=sm_89;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_101,code=sm_101;-gencode;arch=compute_120,code=sm_120 -sage_attention-torch-ext> CMake Warning at /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:146 (message): -sage_attention-torch-ext> Failed to compute shorthash for libnvrtc.so -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> CMake Warning (dev) at /nix/store/0vnarm4qjnj16dr3zj9kwq6bn79c0icn-cmake-3.31.7/share/cmake-3.31/Modules/FindPackageHandleStandardArgs.cmake:441 (message): -sage_attention-torch-ext> The package name passed to `find_package_handle_standard_args` (nvtx3) does -sage_attention-torch-ext> not match the name of the calling package (Caffe2). This can lead to -sage_attention-torch-ext> problems in calling code that expects `find_package` result variables -sage_attention-torch-ext> (e.g., `_FOUND`) to follow a certain pattern. -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:184 (find_package_handle_standard_args) -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> This warning is for project developers. Use -Wno-dev to suppress it. -sage_attention-torch-ext> -sage_attention-torch-ext> -- Could NOT find nvtx3 (missing: nvtx3_dir) -sage_attention-torch-ext> CMake Warning at /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:190 (message): -sage_attention-torch-ext> Cannot find NVTX3, find old NVTX instead -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- USE_CUDNN is set to 0. Compiling without cuDNN support -sage_attention-torch-ext> -- USE_CUSPARSELT is set to 0. Compiling without cuSPARSELt support -sage_attention-torch-ext> -- USE_CUDSS is set to 0. Compiling without cuDSS support -sage_attention-torch-ext> -- USE_CUFILE is set to 0. Compiling without cuFile support -sage_attention-torch-ext> CMake Warning at /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/utils.cmake:328 (message): -sage_attention-torch-ext> In the future we will require one to explicitly pass TORCH_CUDA_ARCH_LIST -sage_attention-torch-ext> to cmake instead of implicitly setting it as an env variable. This will -sage_attention-torch-ext> become a FATAL_ERROR in future version of pytorch. -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:337 (torch_cuda_get_nvcc_gencode_flag) -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- Added CUDA NVCC flags for: -gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_89,code=sm_89;-gencode;arch=compute_90,code=sm_90 -sage_attention-torch-ext> CMake Warning at /nix/store/6drs80sxjhskdki55g5k1dw0jzbd258w-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:22 (message): -sage_attention-torch-ext> static library kineto_LIBRARY-NOTFOUND not found. -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/6drs80sxjhskdki55g5k1dw0jzbd258w-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:125 (append_torchlib_if_found) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> CMake Warning at /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:22 (message): -sage_attention-torch-ext> static library kineto_LIBRARY-NOTFOUND not found. -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/dzz5brlw0xzs9hp3v8fvvwcvkmsr3ls9-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:125 (append_torchlib_if_found) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- Found Torch: /nix/store/mrq1wi2biib2p1mks17g8g5sc4fd492r-python3.13-torch-2.8.0-lib/lib/libtorch.so -sage_attention-torch-ext> -- CUDA target architectures: 7.0;7.5;8.0;8.6;8.9;9.0;10.0;10.1;12.0 -sage_attention-torch-ext> -- CUDA supported target architectures: 7.0;7.5;8.0;8.6;8.9;9.0;10.0;10.1;12.0 -sage_attention-torch-ext> -- PyTorch: Header version is: 12.9 -sage_attention-torch-ext> -- Found Torch: /nix/store/8sicfhvzq84gnxiwybyjgp80pcynamzn-python3.13-torch-2.7.1-lib/lib/libtorch.so -sage_attention-torch-ext> -- CUDA target architectures: 7.0;7.5;8.0;8.6;8.9;9.0 -sage_attention-torch-ext> -- CUDA supported target architectures: 7.0;7.5;8.0;8.6;8.9;9.0 -sage_attention-torch-ext> -- Found Python: /nix/store/aikr517kmcd8r2nrrj70jq71d7352qiq-python3-3.13.6-env/bin/python (found version "3.13.6") found components: Interpreter -sage_attention-torch-ext> CMake Warning at /nix/store/483ma0klnbln74izv5jiyila52bfwqxh-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:146 (message): -sage_attention-torch-ext> Failed to compute shorthash for libnvrtc.so -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/483ma0klnbln74izv5jiyila52bfwqxh-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/483ma0klnbln74izv5jiyila52bfwqxh-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- USE_CUDNN is set to 0. Compiling without cuDNN support -sage_attention-torch-ext> -- USE_CUSPARSELT is set to 0. Compiling without cuSPARSELt support -sage_attention-torch-ext> -- USE_CUDSS is set to 0. Compiling without cuDSS support -sage_attention-torch-ext> -- USE_CUFILE is set to 0. Compiling without cuFile support -sage_attention-torch-ext> -- Added CUDA NVCC flags for: -gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_89,code=sm_89;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_101,code=sm_101;-gencode;arch=compute_120,code=sm_120 -sage_attention-torch-ext> CMake Warning at /nix/store/483ma0klnbln74izv5jiyila52bfwqxh-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:22 (message): -sage_attention-torch-ext> static library kineto_LIBRARY-NOTFOUND not found. -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/483ma0klnbln74izv5jiyila52bfwqxh-python3.13-torch-2.8.0/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:125 (append_torchlib_if_found) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- Found Torch: /nix/store/zccgvlbr93bhyia3sr9f2mddmkp2jyx7-python3.13-torch-2.8.0-lib/lib/libtorch.so -sage_attention-torch-ext> -- CUDA target architectures: 7.0;7.5;8.0;8.6;8.9;9.0;10.0;10.1;12.0 -sage_attention-torch-ext> -- CUDA supported target architectures: 7.0;7.5;8.0;8.6;8.9;9.0;10.0;10.1;12.0 -sage_attention-torch-ext> -- PyTorch: Header version is: 12.8 -sage_attention-torch-ext> -- Found Python: /nix/store/wirj6dihrpcch7idfd7jy4l0hqfsgkk1-python3-3.13.6-env/bin/python (found version "3.13.6") found components: Interpreter -sage_attention-torch-ext> CMake Warning at /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:146 (message): -sage_attention-torch-ext> Failed to compute shorthash for libnvrtc.so -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> CMake Warning (dev) at /nix/store/0vnarm4qjnj16dr3zj9kwq6bn79c0icn-cmake-3.31.7/share/cmake-3.31/Modules/FindPackageHandleStandardArgs.cmake:441 (message): -sage_attention-torch-ext> The package name passed to `find_package_handle_standard_args` (nvtx3) does -sage_attention-torch-ext> not match the name of the calling package (Caffe2). This can lead to -sage_attention-torch-ext> problems in calling code that expects `find_package` result variables -sage_attention-torch-ext> (e.g., `_FOUND`) to follow a certain pattern. -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:184 (find_package_handle_standard_args) -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> This warning is for project developers. Use -Wno-dev to suppress it. -sage_attention-torch-ext> -sage_attention-torch-ext> -- Could NOT find nvtx3 (missing: nvtx3_dir) -sage_attention-torch-ext> CMake Warning at /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:190 (message): -sage_attention-torch-ext> Cannot find NVTX3, find old NVTX instead -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- USE_CUDNN is set to 0. Compiling without cuDNN support -sage_attention-torch-ext> -- USE_CUSPARSELT is set to 0. Compiling without cuSPARSELt support -sage_attention-torch-ext> -- USE_CUDSS is set to 0. Compiling without cuDSS support -sage_attention-torch-ext> -- USE_CUFILE is set to 0. Compiling without cuFile support -sage_attention-torch-ext> CMake Warning at /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/utils.cmake:328 (message): -sage_attention-torch-ext> In the future we will require one to explicitly pass TORCH_CUDA_ARCH_LIST -sage_attention-torch-ext> to cmake instead of implicitly setting it as an env variable. This will -sage_attention-torch-ext> become a FATAL_ERROR in future version of pytorch. -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:337 (torch_cuda_get_nvcc_gencode_flag) -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:86 (include) -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- Added CUDA NVCC flags for: -gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_89,code=sm_89;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_101,code=sm_101;-gencode;arch=compute_120,code=sm_120 -sage_attention-torch-ext> CMake Warning at /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:22 (message): -sage_attention-torch-ext> static library kineto_LIBRARY-NOTFOUND not found. -sage_attention-torch-ext> Call Stack (most recent call first): -sage_attention-torch-ext> /nix/store/4ww34a0xcdm3baaz7y2dnrr38r2yjwwx-python3.13-torch-2.7.1/lib/python3.13/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:125 (append_torchlib_if_found) -sage_attention-torch-ext> CMakeLists.txt:30 (find_package) -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- Found Torch: /nix/store/35sj4in2ddx47klyg96qmkpd4vh8py94-python3.13-torch-2.7.1-lib/lib/libtorch.so -sage_attention-torch-ext> -- CUDA target architectures: 7.0;7.5;8.0;8.6;8.9;9.0;10.0;10.1;12.0 -sage_attention-torch-ext> -- CUDA supported target architectures: 7.0;7.5;8.0;8.6;8.9;9.0;10.0;10.1;12.0 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm80: 8.0 -sage_attention-torch-ext> -- Capabilities for kernel _fused: 9.0a;8.0;8.9 -sage_attention-torch-ext> -- Capabilities for kernel _qattn: 9.0a;8.0;8.9 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm90: 9.0a -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm89: 8.9 -sage_attention-torch-ext> -- Configuring done (9.4s) -sage_attention-torch-ext> -- Generating done (0.0s) -sage_attention-torch-ext> CMake Warning: -sage_attention-torch-ext> Manually-specified variables were not used by the project: -sage_attention-torch-ext> -sage_attention-torch-ext> BUILD_TESTING -sage_attention-torch-ext> CMAKE_EXPORT_NO_PACKAGE_REGISTRY -sage_attention-torch-ext> CMAKE_INSTALL_BINDIR -sage_attention-torch-ext> CMAKE_INSTALL_DOCDIR -sage_attention-torch-ext> CMAKE_INSTALL_INCLUDEDIR -sage_attention-torch-ext> CMAKE_INSTALL_INFODIR -sage_attention-torch-ext> CMAKE_INSTALL_LIBDIR -sage_attention-torch-ext> CMAKE_INSTALL_LIBEXECDIR -sage_attention-torch-ext> CMAKE_INSTALL_LOCALEDIR -sage_attention-torch-ext> CMAKE_INSTALL_MANDIR -sage_attention-torch-ext> CMAKE_INSTALL_SBINDIR -sage_attention-torch-ext> CMAKE_POLICY_DEFAULT_CMP0025 -sage_attention-torch-ext> CUDAToolkit_INCLUDE_DIR -sage_attention-torch-ext> PROTOC_EXE -sage_attention-torch-ext> PYBIND11_PYTHONLIBS_OVERWRITE -sage_attention-torch-ext> PYTHON_EXECUTABLE -sage_attention-torch-ext> PYTHON_INCLUDE_DIR -sage_attention-torch-ext> PYTHON_SITE_PACKAGES -sage_attention-torch-ext> Protobuf_PROTOC_EXE -sage_attention-torch-ext> Protobuf_PROTOC_EXECUTABLE -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- Build files have been written to: /build/source/build -sage_attention-torch-ext> cmake: enabled parallel building -sage_attention-torch-ext> cmake: enabled parallel installing -sage_attention-torch-ext> Running phase: buildPhase -sage_attention-torch-ext> build flags: -j21 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm80: 8.0 -sage_attention-torch-ext> -- Capabilities for kernel _qattn: 9.0a;8.0;8.9 -sage_attention-torch-ext> -- Capabilities for kernel _fused: 9.0a;8.0;8.9 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm89: 8.9 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm90: 9.0a -sage_attention-torch-ext> -- Configuring done (9.5s) -sage_attention-torch-ext> -- Generating done (0.0s) -sage_attention-torch-ext> CMake Warning: -sage_attention-torch-ext> Manually-specified variables were not used by the project: -sage_attention-torch-ext> -sage_attention-torch-ext> BUILD_TESTING -sage_attention-torch-ext> CMAKE_EXPORT_NO_PACKAGE_REGISTRY -sage_attention-torch-ext> CMAKE_INSTALL_BINDIR -sage_attention-torch-ext> CMAKE_INSTALL_DOCDIR -sage_attention-torch-ext> CMAKE_INSTALL_INCLUDEDIR -sage_attention-torch-ext> CMAKE_INSTALL_INFODIR -sage_attention-torch-ext> CMAKE_INSTALL_LIBDIR -sage_attention-torch-ext> CMAKE_INSTALL_LIBEXECDIR -sage_attention-torch-ext> CMAKE_INSTALL_LOCALEDIR -sage_attention-torch-ext> CMAKE_INSTALL_MANDIR -sage_attention-torch-ext> CMAKE_INSTALL_SBINDIR -sage_attention-torch-ext> CMAKE_POLICY_DEFAULT_CMP0025 -sage_attention-torch-ext> CUDAToolkit_INCLUDE_DIR -sage_attention-torch-ext> PROTOC_EXE -sage_attention-torch-ext> PYBIND11_PYTHONLIBS_OVERWRITE -sage_attention-torch-ext> PYTHON_EXECUTABLE -sage_attention-torch-ext> PYTHON_INCLUDE_DIR -sage_attention-torch-ext> PYTHON_SITE_PACKAGES -sage_attention-torch-ext> Protobuf_PROTOC_EXE -sage_attention-torch-ext> Protobuf_PROTOC_EXECUTABLE -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- Build files have been written to: /build/source/build -sage_attention-torch-ext> cmake: enabled parallel building -sage_attention-torch-ext> cmake: enabled parallel installing -sage_attention-torch-ext> Running phase: buildPhase -sage_attention-torch-ext> build flags: -j21 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm89: 8.9 -sage_attention-torch-ext> -- Capabilities for kernel _fused: 9.0a;8.0;8.9 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm90: 9.0a -sage_attention-torch-ext> -- Capabilities for kernel _qattn: 9.0a;8.0;8.9 -sage_attention-torch-ext> -- Capabilities for kernel _qattn: 9.0a;8.0;8.9 -sage_attention-torch-ext> -- Capabilities for kernel _fused: 9.0a;8.0;8.9 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm80: 8.0 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm89: 8.9 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm80: 8.0 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm90: 9.0a -sage_attention-torch-ext> -- Configuring done (9.5s) -sage_attention-torch-ext> -- Configuring done (9.6s) -sage_attention-torch-ext> -- Generating done (0.0s) -sage_attention-torch-ext> CMake Warning: -sage_attention-torch-ext> Manually-specified variables were not used by the project: -sage_attention-torch-ext> -sage_attention-torch-ext> BUILD_TESTING -sage_attention-torch-ext> CMAKE_EXPORT_NO_PACKAGE_REGISTRY -sage_attention-torch-ext> CMAKE_INSTALL_BINDIR -sage_attention-torch-ext> CMAKE_INSTALL_DOCDIR -sage_attention-torch-ext> CMAKE_INSTALL_INCLUDEDIR -sage_attention-torch-ext> CMAKE_INSTALL_INFODIR -sage_attention-torch-ext> CMAKE_INSTALL_LIBDIR -sage_attention-torch-ext> CMAKE_INSTALL_LIBEXECDIR -sage_attention-torch-ext> CMAKE_INSTALL_LOCALEDIR -sage_attention-torch-ext> CMAKE_INSTALL_MANDIR -sage_attention-torch-ext> CMAKE_INSTALL_SBINDIR -sage_attention-torch-ext> CMAKE_POLICY_DEFAULT_CMP0025 -sage_attention-torch-ext> CUDAToolkit_INCLUDE_DIR -sage_attention-torch-ext> PROTOC_EXE -sage_attention-torch-ext> PYBIND11_PYTHONLIBS_OVERWRITE -sage_attention-torch-ext> PYTHON_EXECUTABLE -sage_attention-torch-ext> PYTHON_INCLUDE_DIR -sage_attention-torch-ext> PYTHON_SITE_PACKAGES -sage_attention-torch-ext> Protobuf_PROTOC_EXE -sage_attention-torch-ext> Protobuf_PROTOC_EXECUTABLE -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- Build files have been written to: /build/source/build -sage_attention-torch-ext> -- Generating done (0.0s) -sage_attention-torch-ext> CMake Warning: -sage_attention-torch-ext> Manually-specified variables were not used by the project: -sage_attention-torch-ext> -sage_attention-torch-ext> BUILD_TESTING -sage_attention-torch-ext> CMAKE_EXPORT_NO_PACKAGE_REGISTRY -sage_attention-torch-ext> CMAKE_INSTALL_BINDIR -sage_attention-torch-ext> CMAKE_INSTALL_DOCDIR -sage_attention-torch-ext> CMAKE_INSTALL_INCLUDEDIR -sage_attention-torch-ext> CMAKE_INSTALL_INFODIR -sage_attention-torch-ext> CMAKE_INSTALL_LIBDIR -sage_attention-torch-ext> CMAKE_INSTALL_LIBEXECDIR -sage_attention-torch-ext> CMAKE_INSTALL_LOCALEDIR -sage_attention-torch-ext> CMAKE_INSTALL_MANDIR -sage_attention-torch-ext> CMAKE_INSTALL_SBINDIR -sage_attention-torch-ext> CMAKE_POLICY_DEFAULT_CMP0025 -sage_attention-torch-ext> CUDAToolkit_INCLUDE_DIR -sage_attention-torch-ext> PROTOC_EXE -sage_attention-torch-ext> PYBIND11_PYTHONLIBS_OVERWRITE -sage_attention-torch-ext> PYTHON_EXECUTABLE -sage_attention-torch-ext> PYTHON_INCLUDE_DIR -sage_attention-torch-ext> PYTHON_SITE_PACKAGES -sage_attention-torch-ext> Protobuf_PROTOC_EXE -sage_attention-torch-ext> Protobuf_PROTOC_EXECUTABLE -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- Build files have been written to: /build/source/build -sage_attention-torch-ext> cmake: enabled parallel building -sage_attention-torch-ext> cmake: enabled parallel installing -sage_attention-torch-ext> Running phase: buildPhase -sage_attention-torch-ext> build flags: -j21 -sage_attention-torch-ext> cmake: enabled parallel building -sage_attention-torch-ext> cmake: enabled parallel installing -sage_attention-torch-ext> Running phase: buildPhase -sage_attention-torch-ext> build flags: -j21 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm89: 8.9 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm80: 8.0 -sage_attention-torch-ext> -- Capabilities for kernel _fused: 9.0a;8.0;8.9 -sage_attention-torch-ext> -- Capabilities for kernel _qattn_sm90: 9.0a -sage_attention-torch-ext> -- Capabilities for kernel _qattn: 9.0a;8.0;8.9 -sage_attention-torch-ext> -- Configuring done (9.8s) -sage_attention-torch-ext> -- Generating done (0.0s) -sage_attention-torch-ext> CMake Warning: -sage_attention-torch-ext> Manually-specified variables were not used by the project: -sage_attention-torch-ext> -sage_attention-torch-ext> BUILD_TESTING -sage_attention-torch-ext> CMAKE_EXPORT_NO_PACKAGE_REGISTRY -sage_attention-torch-ext> CMAKE_INSTALL_BINDIR -sage_attention-torch-ext> CMAKE_INSTALL_DOCDIR -sage_attention-torch-ext> CMAKE_INSTALL_INCLUDEDIR -sage_attention-torch-ext> CMAKE_INSTALL_INFODIR -sage_attention-torch-ext> CMAKE_INSTALL_LIBDIR -sage_attention-torch-ext> CMAKE_INSTALL_LIBEXECDIR -sage_attention-torch-ext> CMAKE_INSTALL_LOCALEDIR -sage_attention-torch-ext> CMAKE_INSTALL_MANDIR -sage_attention-torch-ext> CMAKE_INSTALL_SBINDIR -sage_attention-torch-ext> CMAKE_POLICY_DEFAULT_CMP0025 -sage_attention-torch-ext> CUDAToolkit_INCLUDE_DIR -sage_attention-torch-ext> PROTOC_EXE -sage_attention-torch-ext> PYBIND11_PYTHONLIBS_OVERWRITE -sage_attention-torch-ext> PYTHON_EXECUTABLE -sage_attention-torch-ext> PYTHON_INCLUDE_DIR -sage_attention-torch-ext> PYTHON_SITE_PACKAGES -sage_attention-torch-ext> Protobuf_PROTOC_EXE -sage_attention-torch-ext> Protobuf_PROTOC_EXECUTABLE -sage_attention-torch-ext> -sage_attention-torch-ext> -sage_attention-torch-ext> -- Build files have been written to: /build/source/build -sage_attention-torch-ext> cmake: enabled parallel building -sage_attention-torch-ext> cmake: enabled parallel installing -sage_attention-torch-ext> Running phase: buildPhase -sage_attention-torch-ext> build flags: -j21 -sage_attention-torch-ext> [1/12] Building CXX object CMakeFiles/_sage_attention_1369690_dirty.dir/torch-ext/torch_binding.cpp.o -sage_attention-torch-ext> [1/12] Building CXX object CMakeFiles/_sage_attention_1369690_dirty.dir/torch-ext/torch_binding.cpp.o -sage_attention-torch-ext> [1/12] Building CXX object CMakeFiles/_sage_attention_1369690_dirty.dir/torch-ext/torch_binding.cpp.o -sage_attention-torch-ext> [1/12] Building CXX object CMakeFiles/_sage_attention_1369690_dirty.dir/torch-ext/torch_binding.cpp.o -sage_attention-torch-ext> [1/12] Building CXX object CMakeFiles/_sage_attention_1369690_dirty.dir/torch-ext/torch_binding.cpp.o -sage_attention-torch-ext> [2/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 264 bytes stack frame, 268 bytes spill stores, 280 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 264 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 264 bytes stack frame, 268 bytes spill stores, 280 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 264 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 288 bytes stack frame, 260 bytes spill stores, 268 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 288 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 288 bytes stack frame, 260 bytes spill stores, 268 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 288 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 264 bytes spill stores, 264 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 264 bytes spill stores, 264 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 256 bytes spill stores, 260 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 256 bytes spill stores, 260 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 296 bytes spill stores, 300 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 296 bytes spill stores, 300 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 232 bytes spill stores, 248 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 232 bytes spill stores, 248 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 292 bytes spill stores, 292 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 292 bytes spill stores, 292 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 212 bytes spill stores, 224 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 212 bytes spill stores, 224 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 228 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 228 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [2/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 547.358 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 535.674 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 543.831 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 544.308 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 84 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 543.627 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 84 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 546.697 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 545.861 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 542.011 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 457.024 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 453.257 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 573.264 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 574.183 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 36 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 452.252 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 36 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 450.873 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 581.817 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 566.486 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 285.493 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 286.733 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 252.970 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 253.284 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 292.503 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 291.856 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 252.089 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.094 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 306.229 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 306.386 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 312.993 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 314.081 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 307.194 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 307.151 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 310.611 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 311.148 ms -sage_attention-torch-ext> [2/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(627): warning #177-D: variable "padded_kv_len" was declared but never referenced -sage_attention-torch-ext> int qo_len, kv_len, padded_kv_len, num_qo_heads, num_kv_heads; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> Remark: The warnings can be suppressed with "-diag-suppress " -sage_attention-torch-ext> -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(803): warning #177-D: variable "padded_kv_len" was declared but never referenced -sage_attention-torch-ext> int qo_len, kv_len, padded_kv_len, num_qo_heads, num_kv_heads; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(170): warning #177-D: variable "sO" was declared but never referenced -sage_attention-torch-ext> half *sO = (half*)smem_; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> ptxas info : 10 bytes gmem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> [3/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 12 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 469.625 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 12 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 463.912 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 571.529 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 569.671 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 561.608 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 561.001 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 556.035 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 557.200 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 479.684 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 481.712 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 56 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 603.697 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 56 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 598.014 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 480.607 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 477.343 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 590.169 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 581.773 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 234.406 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 243.416 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.702 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.739 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 233.883 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 234.683 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.138 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.912 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 257.751 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 257.498 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 263.645 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 262.543 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 256.486 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 256.154 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 260.577 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 261.259 ms -sage_attention-torch-ext> [4/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(627): warning #177-D: variable "padded_kv_len" was declared but never referenced -sage_attention-torch-ext> int qo_len, kv_len, padded_kv_len, num_qo_heads, num_kv_heads; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> Remark: The warnings can be suppressed with "-diag-suppress " -sage_attention-torch-ext> -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(803): warning #177-D: variable "padded_kv_len" was declared but never referenced -sage_attention-torch-ext> int qo_len, kv_len, padded_kv_len, num_qo_heads, num_kv_heads; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(170): warning #177-D: variable "sO" was declared but never referenced -sage_attention-torch-ext> half *sO = (half*)smem_; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> ptxas info : 11 bytes gmem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 170.445 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 170.922 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 178.615 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 173.053 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 167.675 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 169.070 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 173.540 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 172.860 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 176.685 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 174.977 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 181.300 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 179.394 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 173.679 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 174.799 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 179.923 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 180.588 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.492 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 105.720 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.093 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.763 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.682 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.290 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.730 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 110.856 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 111.653 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 113.112 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 117.481 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 116.677 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 112.242 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 112.828 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 116.094 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 116.392 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 156.453 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 156.250 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 161.338 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 160.956 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 155.991 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 158.280 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 160.362 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 160.147 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 163.361 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 166.793 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 169.915 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 168.781 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 164.936 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 163.500 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 169.078 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 170.510 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 104.161 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 104.761 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.125 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 108.052 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 104.397 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 104.650 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.489 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 108.521 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 110.758 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 110.970 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 117.195 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 123.621 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 112.699 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 110.730 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 114.895 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 113.441 ms -sage_attention-torch-ext> [3/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 68 bytes spill stores, 52 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 68 bytes spill stores, 52 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 84 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 84 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 20 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 20 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 36 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 36 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [2/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(627): warning #177-D: variable "padded_kv_len" was declared but never referenced -sage_attention-torch-ext> int qo_len, kv_len, padded_kv_len, num_qo_heads, num_kv_heads; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> Remark: The warnings can be suppressed with "-diag-suppress " -sage_attention-torch-ext> -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(803): warning #177-D: variable "padded_kv_len" was declared but never referenced -sage_attention-torch-ext> int qo_len, kv_len, padded_kv_len, num_qo_heads, num_kv_heads; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(170): warning #177-D: variable "sO" was declared but never referenced -sage_attention-torch-ext> half *sO = (half*)smem_; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> ptxas info : 11 bytes gmem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 165.854 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 165.334 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 171.384 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 168.936 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 163.754 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 164.901 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 168.686 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 168.149 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 172.498 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 169.288 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 176.070 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 174.334 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 168.530 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 170.021 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 175.010 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 178.739 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 103.287 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 102.746 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.637 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 107.102 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 103.971 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 103.396 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.978 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.550 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 108.782 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 108.261 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 112.849 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 114.214 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.221 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.147 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 112.909 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 112.612 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 151.667 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 153.445 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 156.690 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 156.319 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 151.849 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 152.070 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 156.359 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 156.937 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 159.221 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 159.340 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 164.004 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 164.129 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 161.468 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 159.516 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 164.971 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 165.774 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 101.825 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 101.679 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 107.124 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 107.857 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 101.784 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 101.641 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 105.927 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 104.587 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 107.968 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 108.193 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 112.581 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 113.614 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.781 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 107.417 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.877 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.774 ms -sage_attention-torch-ext> [4/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 24 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 24 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 88 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 88 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 24 bytes stack frame, 36 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 24 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 24 bytes stack frame, 36 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 24 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [5/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_fuse_v_mean_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 826.557 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 816.161 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 828.861 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 826.193 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 816.337 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 808.415 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 801.612 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 815.667 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 707.437 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 714.530 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 48 bytes spill stores, 52 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1001.111 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 48 bytes spill stores, 52 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1010.184 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 767.697 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 787.498 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 966.520 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1003.302 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 245 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 398.364 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 245 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 412.057 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 438.811 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 452.307 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 444.824 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 438.867 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 417.958 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 437.697 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 469.650 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 462.005 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 467.912 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 464.076 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 459.537 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 447.576 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 451.094 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 443.545 ms -sage_attention-torch-ext> [3/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_fuse_v_scale_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 72 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 590.184 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 72 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 587.826 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 596.714 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 602.905 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 580.377 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 576.021 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 585.040 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 586.444 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 168 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 525.571 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 168 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 525.233 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 628.043 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 627.475 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 503.080 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 511.243 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 960.108 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1038.392 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 251 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 397.404 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 251 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 411.247 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 418.848 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 424.521 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 404.732 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 407.096 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 424.318 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 415.598 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 451.793 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 447.662 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 458.394 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 456.104 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 444.661 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 444.848 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 459.398 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 464.293 ms -sage_attention-torch-ext> [6/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/fused/fused.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 54.421 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 47.949 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 40.944 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 39.316 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.232 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 6.566 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.159 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.136 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 7.061 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 6.939 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 5.816 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 5.689 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.363 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.382 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.423 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.503 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.361 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.247 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.299 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.188 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 19.787 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 14.026 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 13.787 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 14.131 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 19.009 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 13.657 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 13.523 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 13.761 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 17.134 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.366 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.538 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.357 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 17.425 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.341 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.212 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.247 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 18.561 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 13.023 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.779 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.803 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 17.507 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.371 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.707 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.546 ms -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 36.806 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 34.855 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 27.632 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 27.810 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 5.357 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 5.366 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 5.246 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 5.272 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 5.183 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 4.962 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 4.507 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 4.381 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.430 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.333 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.380 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.435 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.494 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 12.143 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.615 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.365 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 17.454 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 13.094 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 13.174 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 13.068 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 17.697 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 13.099 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 13.042 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 12.930 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 14.211 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.913 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.271 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.302 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.686 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.103 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.086 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.123 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.082 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.140 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.702 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.294 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.023 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.350 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.314 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.240 ms -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 39 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 29.520 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 39 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 28.105 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 22.644 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 23.426 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.446 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.349 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.296 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.317 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.316 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.230 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 18 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.816 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 18 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.882 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.155 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.172 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.182 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.123 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.209 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.742 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.336 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.164 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.758 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.309 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.158 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.174 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.852 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.224 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.222 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.196 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.587 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.197 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.321 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.244 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.772 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.255 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.304 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.249 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.109 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.325 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.350 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.384 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.139 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.422 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.451 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.282 ms -sage_attention-torch-ext> [3/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 68 bytes spill stores, 52 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 68 bytes spill stores, 52 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 84 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 84 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 20 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 20 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 36 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 36 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [7/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_fuse_v_scale_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 72 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 999.385 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 72 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1008.310 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1010.558 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1022.908 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 993.110 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 986.279 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 894.060 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 598.193 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 168 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 539.079 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 168 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 525.486 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 636.814 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 643.895 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 515.792 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 516.644 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 636.165 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 633.284 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 251 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 243.158 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 251 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.717 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 256.472 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 251.914 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.256 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.694 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 256.876 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 254.324 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 271.323 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 268.801 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 279.458 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 276.534 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 269.084 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 270.328 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 279.058 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 275.557 ms -sage_attention-torch-ext> [5/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_fuse_v_scale_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 264 bytes stack frame, 276 bytes spill stores, 280 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 264 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 264 bytes stack frame, 276 bytes spill stores, 280 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 264 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 288 bytes stack frame, 268 bytes spill stores, 268 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 288 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 288 bytes stack frame, 268 bytes spill stores, 268 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 288 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 272 bytes spill stores, 276 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 272 bytes spill stores, 276 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 264 bytes spill stores, 264 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 264 bytes spill stores, 264 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 224 bytes stack frame, 252 bytes spill stores, 248 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 224 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 224 bytes stack frame, 252 bytes spill stores, 248 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 224 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 216 bytes spill stores, 232 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 216 bytes spill stores, 232 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 216 bytes stack frame, 244 bytes spill stores, 236 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 216 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 216 bytes stack frame, 244 bytes spill stores, 236 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 216 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 212 bytes spill stores, 224 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 212 bytes spill stores, 224 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 238 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 238 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [4/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_fuse_v_scale_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 264 bytes stack frame, 276 bytes spill stores, 280 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 264 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 264 bytes stack frame, 276 bytes spill stores, 280 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 264 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 288 bytes stack frame, 268 bytes spill stores, 268 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 288 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 288 bytes stack frame, 268 bytes spill stores, 268 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 288 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 272 bytes spill stores, 276 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 272 bytes spill stores, 276 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 264 bytes spill stores, 264 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 264 bytes spill stores, 264 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 224 bytes stack frame, 252 bytes spill stores, 248 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 224 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 224 bytes stack frame, 252 bytes spill stores, 248 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 224 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 216 bytes spill stores, 232 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 216 bytes spill stores, 232 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 216 bytes stack frame, 244 bytes spill stores, 236 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 216 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 216 bytes stack frame, 244 bytes spill stores, 236 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 216 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 212 bytes spill stores, 224 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 212 bytes spill stores, 224 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 238 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 238 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [4/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 620.508 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 616.107 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 772.153 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 774.453 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 782.307 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 766.628 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 763.804 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 774.962 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 655.297 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 658.864 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 818.494 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 834.707 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 674.744 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 694.457 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 934.425 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 931.353 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 354.255 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 343.199 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 370.905 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 399.378 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 334.581 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 375.615 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 359.839 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 410.376 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 411.621 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 383.662 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 439.704 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 437.275 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 247 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 446.328 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 247 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 437.873 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 430.902 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 298.938 ms -sage_attention-torch-ext> [5/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 8 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 8 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [5/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 12 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 790.771 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 12 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 792.881 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 980.108 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 977.646 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 966.776 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 851.252 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 556.777 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 558.364 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 490.315 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 521.780 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 56 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 619.202 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 56 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 611.598 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 488.636 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 495.391 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 603.818 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 603.052 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 243.290 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.699 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.971 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.242 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 244.108 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 239.020 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.400 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 244.384 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 271.587 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 262.148 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 271.074 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 264.960 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 262.943 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 268.571 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 300.543 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 458.510 ms -sage_attention-torch-ext> [6/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_fuse_v_mean_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 20 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 20 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 245 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 245 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [6/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 998.552 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 826.635 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 76 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 615.748 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 76 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 606.650 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 603.297 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 593.839 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 603.450 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 602.685 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 501.419 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 496.502 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 632.978 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 631.913 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 499.363 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 497.475 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 624.370 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 628.275 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 317.593 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 337.409 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 338.980 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 325.425 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 321.102 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 330.768 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 332.775 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 325.485 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 341.080 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 339.245 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 346.842 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 346.461 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 344.575 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 338.607 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 346.476 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 345.178 ms -sage_attention-torch-ext> [8/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 749.212 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 735.567 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 898.978 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 918.429 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 914.085 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 895.687 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 864.867 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 914.954 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 793.578 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 786.775 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1000.400 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 962.812 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 790.596 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 780.630 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 884.545 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 915.802 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 379.958 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 386.163 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 391.533 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 386.332 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 377.969 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 375.346 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 388.152 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 387.354 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 418.584 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 416.620 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 422.887 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 418.367 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 247 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 412.537 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 247 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 412.410 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 420.907 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 422.828 ms -sage_attention-torch-ext> [9/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 968.796 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 997.981 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 76 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1003.385 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 76 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 976.590 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 986.321 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 988.531 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 960.047 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 983.177 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 841.694 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 835.789 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1054.160 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1071.933 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 809.331 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 815.178 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 792.847 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 606.934 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 310.416 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 305.370 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 313.186 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 315.267 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 332.934 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 319.704 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 310.698 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 314.573 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 327.737 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 326.537 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 332.464 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 333.084 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 327.492 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 326.340 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 333.489 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 334.503 ms -sage_attention-torch-ext> [6/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 264 bytes stack frame, 268 bytes spill stores, 280 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 264 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 264 bytes stack frame, 268 bytes spill stores, 280 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 264 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 288 bytes stack frame, 260 bytes spill stores, 268 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 288 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 288 bytes stack frame, 260 bytes spill stores, 268 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 288 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 264 bytes spill stores, 264 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 264 bytes spill stores, 264 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 256 bytes spill stores, 260 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 256 bytes spill stores, 260 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 296 bytes spill stores, 300 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 296 bytes spill stores, 300 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 232 bytes spill stores, 248 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 232 bytes spill stores, 248 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 292 bytes spill stores, 292 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 272 bytes stack frame, 292 bytes spill stores, 292 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 272 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 212 bytes spill stores, 224 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 256 bytes stack frame, 212 bytes spill stores, 224 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 256 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 228 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 228 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 242 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [7/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 88 bytes stack frame, 156 bytes spill stores, 148 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 88 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 581.614 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 88 bytes stack frame, 156 bytes spill stores, 148 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 88 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 579.645 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 96 bytes spill stores, 92 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 566.773 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 96 bytes spill stores, 92 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 824.023 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 32 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 943.343 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 32 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 951.713 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 948.674 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 943.709 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 84 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 839.100 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 84 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 832.107 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1057.496 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1076.174 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 164 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 870.170 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 164 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 858.787 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 136 bytes spill stores, 128 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1044.382 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 136 bytes spill stores, 128 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1058.303 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 399.733 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 391.304 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 406.681 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 402.870 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 396.301 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 387.645 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 401.033 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 401.514 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 436.194 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 426.740 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 437.151 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 437.547 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 428.190 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 432.674 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 441.354 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 441.692 ms -sage_attention-torch-ext> [7/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 247 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 247 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [8/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 8 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 8 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [8/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_fuse_v_mean_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 876.597 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 972.582 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 987.989 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 995.288 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 994.117 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 982.604 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 981.111 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 970.839 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 846.454 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 849.497 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 48 bytes spill stores, 52 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1031.414 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 48 bytes spill stores, 52 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1048.652 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 835.362 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 829.827 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1017.734 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1023.163 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 245 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 414.881 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 245 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.112 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 250.384 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 252.290 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.246 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.309 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 252.676 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.916 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 270.735 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 266.398 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 269.132 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 389.197 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 434.975 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 446.924 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 455.440 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 458.558 ms -sage_attention-torch-ext> [9/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_fuse_v_mean_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 20 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 20 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 24 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 245 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 245 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [7/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 24 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 24 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 88 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 88 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 24 bytes stack frame, 36 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 24 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 24 bytes stack frame, 36 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 24 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [10/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 88 bytes stack frame, 156 bytes spill stores, 148 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 88 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 598.749 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 88 bytes stack frame, 156 bytes spill stores, 148 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 88 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 596.165 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 96 bytes spill stores, 92 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 501.088 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 96 bytes spill stores, 92 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 497.756 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 32 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 573.002 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 32 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 581.095 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 588.939 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 584.204 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 84 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 513.593 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 84 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 510.562 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 656.214 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 643.416 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 164 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 527.443 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 164 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 525.280 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 136 bytes spill stores, 128 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 662.868 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 136 bytes spill stores, 128 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 655.125 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 243.729 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 239.722 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.248 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 248.027 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.174 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 239.441 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 257.898 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 250.742 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 268.427 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 263.558 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 268.852 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 272.929 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 261.214 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 264.042 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 274.820 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 274.232 ms -sage_attention-torch-ext> [8/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 237 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 235 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 230 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 232 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 247 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 247 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [9/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/fused/fused.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 53.435 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 47.765 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 41.736 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 40.130 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.070 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 6.856 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 6.834 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 6.975 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 6.991 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 6.724 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 5.817 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 5.687 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.931 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.173 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.894 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.015 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.008 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.857 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.741 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.547 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 19.737 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 13.572 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 14.067 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 13.535 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 19.165 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 13.529 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.434 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 13.280 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 16.255 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.843 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.201 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.213 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 15.513 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.053 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.232 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.991 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 18.070 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.546 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.369 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.080 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 16.883 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.496 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.565 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 12.891 ms -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 36.100 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 33.911 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 26.779 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 27.032 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 5.253 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 4.779 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 4.743 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 5.058 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 4.976 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 4.813 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 4.293 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 4.116 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.428 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.336 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.905 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.222 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.267 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.589 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.818 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.261 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 17.417 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 12.709 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 12.789 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 12.587 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 17.321 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 12.960 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 12.834 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 13.048 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 15.219 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.313 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.251 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.574 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 15.609 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.386 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.207 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.373 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 15.991 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.489 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.332 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.467 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 16.061 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.227 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.531 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 11.228 ms -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 39 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 29.465 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 39 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 27.737 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 22.151 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 22.760 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.458 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.357 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.243 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.209 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.185 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.101 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 18 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.695 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 18 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.691 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.151 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.375 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.427 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.392 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.428 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.915 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.532 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.538 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.748 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.055 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.089 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.733 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.817 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.169 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.169 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.111 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.587 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.140 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.164 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.172 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.131 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.202 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.163 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.177 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.982 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.300 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.334 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.348 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.073 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.380 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.375 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.222 ms -sage_attention-torch-ext> [9/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(627): warning #177-D: variable "padded_kv_len" was declared but never referenced -sage_attention-torch-ext> int qo_len, kv_len, padded_kv_len, num_qo_heads, num_kv_heads; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> Remark: The warnings can be suppressed with "-diag-suppress " -sage_attention-torch-ext> -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(803): warning #177-D: variable "padded_kv_len" was declared but never referenced -sage_attention-torch-ext> int qo_len, kv_len, padded_kv_len, num_qo_heads, num_kv_heads; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(170): warning #177-D: variable "sO" was declared but never referenced -sage_attention-torch-ext> half *sO = (half*)smem_; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> ptxas info : 10 bytes gmem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> [10/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/fused/fused.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 39 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 39 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 18 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 18 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> [10/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/fused/fused.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 39 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 39 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 18 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 18 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> [10/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 926.291 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 988.834 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 998.212 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 981.730 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 84 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1010.516 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 84 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 970.592 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 983.619 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 950.009 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 732.815 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 791.506 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1026.285 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 1020.836 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 36 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 800.657 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 36 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 788.173 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 996.209 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 996.247 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 501.649 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 504.874 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 434.974 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 430.997 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 500.588 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 489.681 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 434.422 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 433.981 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 542.961 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 539.602 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 544.148 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 551.551 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 540.760 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 538.659 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 534.637 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 520.703 ms -sage_attention-torch-ext> [2/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(627): warning #177-D: variable "padded_kv_len" was declared but never referenced -sage_attention-torch-ext> int qo_len, kv_len, padded_kv_len, num_qo_heads, num_kv_heads; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> Remark: The warnings can be suppressed with "-diag-suppress " -sage_attention-torch-ext> -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(803): warning #177-D: variable "padded_kv_len" was declared but never referenced -sage_attention-torch-ext> int qo_len, kv_len, padded_kv_len, num_qo_heads, num_kv_heads; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> /build/source/sage_attention/qattn/qk_int_sv_f8_cuda_sm90.cu(170): warning #177-D: variable "sO" was declared but never referenced -sage_attention-torch-ext> half *sO = (half*)smem_; -sage_attention-torch-ext> ^ -sage_attention-torch-ext> -sage_attention-torch-ext> ptxas info : 28 bytes gmem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 164.629 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 162.500 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 168.109 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 168.494 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 163.334 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 162.936 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 167.919 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 167.494 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 171.728 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 172.263 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 171.204 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 170.337 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 164.521 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 164.146 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 169.215 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 169.211 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 103.110 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 103.017 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.913 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.820 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 103.073 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 102.760 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.418 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.151 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 107.918 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.250 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 113.272 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 113.196 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.622 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.072 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 113.129 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb1EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 112.985 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 151.866 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 152.247 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 157.337 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 156.475 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 152.326 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 151.880 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 156.948 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 157.435 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 159.109 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 159.365 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 164.546 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 165.551 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 160.853 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 167 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 162.161 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 162.325 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj128EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 168 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 161.610 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 100.696 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 100.100 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 103.793 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 103.643 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 99.585 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 100.444 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 104.224 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode0ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 103.844 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.050 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.060 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 110.421 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity3ELS0_3E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.914 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.169 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb0ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 106.104 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E13__nv_bfloat16L8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.643 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int8_sv_f8_attn_kernelILj64ELj128ELj128ELj64EL16QuantGranularity2ELS0_2E6__halfL8MaskMode1ELb1ELb0EEv14CUtensorMap_stS3_S3_PfS4_S4_PT5_S4_jjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 128 registers, used 1 barriers, 16 bytes cumulative stack size, 128 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 109.592 ms -sage_attention-torch-ext> [11/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/qk_int_sv_f16_cuda_sm80.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 227 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 217 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 202 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 202 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 194 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 194 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 185 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 185 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 246 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 246 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 176 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 176 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 184 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 184 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 180 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 180 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 178 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 178 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 201 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 207 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 204 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 204 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 60 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 60 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 68 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 68 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 56 bytes spill stores, 60 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 56 bytes spill stores, 60 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 60 bytes spill stores, 64 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 60 bytes spill stores, 64 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [12/12] Linking CXX shared module _sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> buildPhase completed in 3 minutes 28 seconds -sage_attention-torch-ext> Running phase: installPhase -sage_attention-torch-ext> install flags: -j21 install -sage_attention-torch-ext> [0/1] Install the project... -sage_attention-torch-ext> -- Install configuration: "Release" -sage_attention-torch-ext> -- Installing: /nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/_sage_attention_1369690_dirty/_sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> Running phase: fixupPhase -sage_attention-torch-ext> shrinking RPATHs of ELF executables and libraries in /nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext -sage_attention-torch-ext> shrinking /nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext/sage_attention/_sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> checking for references to /build/ in /nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext... -sage_attention-torch-ext> patching script interpreter paths in /nix/store/bgkz7fkw3xnhpn3s9xl0mh6hzw50igaz-sage_attention-torch-ext -sage_attention-torch-ext> Running phase: installCheckPhase -sage_attention-torch-ext> no Makefile or custom installCheckPhase, doing nothing -sage_attention-torch-ext> Checking of ABI compatibility -sage_attention-torch-ext> 🐍 Checking for compatibility with manylinux_2_28 and Python ABI version 3.9 -sage_attention-torch-ext> ✅ No compatibility issues found -sage_attention-torch-ext> Checking loading kernel with get_kernel -sage_attention-torch-ext> Check whether the kernel can be loaded with get-kernel: sage_attention -sage_attention-torch-ext> [3/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 28 bytes gmem, 224 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 436.735 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 431.396 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 533.219 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 530.670 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 521.712 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 526.479 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 523.878 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 518.861 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 460.431 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 460.510 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 554.292 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 554.013 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 456.477 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 457.482 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 554.059 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 554.122 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 226.781 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 229.174 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 238 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 234.113 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 238 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 234.133 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 236 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 226.809 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 236 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 227.034 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 229 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 229.475 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 229 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 229.017 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 245.178 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 245.310 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 251.848 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 252.673 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 243 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 245.746 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 243 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.534 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 250.643 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 250.444 ms -sage_attention-torch-ext> [4/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_fuse_v_mean_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 28 bytes gmem, 224 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 559.443 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 554.409 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 568.894 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 567.309 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 555.844 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 554.456 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 550.486 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 549.836 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 486.794 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 36 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 487.416 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 48 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 592.195 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 48 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 591.013 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 484.986 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 489.331 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 590.586 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 585.963 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 245 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 239.921 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 245 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.653 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.697 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.332 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 236 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.879 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 236 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.201 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.318 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.491 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 267.306 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 234 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 263.643 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 270.400 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 268.505 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 238 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 263.997 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 238 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 262.643 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 269.183 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb1ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 267.816 ms -sage_attention-torch-ext> [5/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 28 bytes gmem, 224 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 532.802 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 524.131 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 532.873 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 12 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 531.895 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 84 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 529.212 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 84 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 530.045 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 530.147 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 529.409 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 453.739 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 64 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 451.804 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 565.533 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 566.111 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 36 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 451.967 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 36 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 450.567 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 559.371 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 557.747 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 279.939 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 277.584 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.335 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.460 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 281.817 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 281.227 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.478 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 245.385 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 303.254 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 302.164 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 308.583 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 309.322 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 299.838 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 299.171 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 304.632 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 303.234 ms -sage_attention-torch-ext> [6/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 28 bytes gmem, 224 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 573.985 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 566.523 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 76 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 569.328 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 76 bytes spill stores, 44 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 570.085 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 561.603 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 28 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 560.633 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 566.286 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 16 bytes spill stores, 12 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 564.440 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 476.978 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 478.590 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 591.063 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 594.559 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 481.319 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 52 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 477.812 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 590.875 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 594.283 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 300.753 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 301.565 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 308.950 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 308.883 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 300.010 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 296.211 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 305.394 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 307.656 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 325.958 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 321.870 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 328.642 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 329.535 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 321.379 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 322.064 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 329.599 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 328.798 ms -sage_attention-torch-ext> [7/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_fuse_v_scale_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 28 bytes gmem, 224 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 72 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 581.629 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 72 bytes spill stores, 56 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 581.290 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 589.654 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 592.633 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 568.978 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 16 bytes spill stores, 16 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 563.291 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 575.280 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 28 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 574.299 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 168 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 523.598 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 168 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 517.873 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 619.219 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 623.755 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 504.730 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 80 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 506.378 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 613.005 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 60 bytes spill stores, 48 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 607.654 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 251 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 239.654 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 251 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 244.360 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 251.023 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 248.952 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.613 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 243.398 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.045 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.971 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 265.475 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 265.001 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 270.604 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 269.738 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 264.063 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 265.972 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 270.710 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 270.725 ms -sage_attention-torch-ext> [11/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/qk_int_sv_f16_cuda_sm80.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 10 bytes gmem, 80 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 227 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 217 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 202 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 202 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 194 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 194 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 185 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 185 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 246 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 246 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 176 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 176 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 184 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 184 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 180 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 180 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 178 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 178 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 201 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 207 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 204 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 204 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 60 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 60 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 68 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 68 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 56 bytes spill stores, 60 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 56 bytes spill stores, 60 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 60 bytes spill stores, 64 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 60 bytes spill stores, 64 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> [11/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/qk_int_sv_f16_cuda_sm80.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 562.853 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 527.005 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 568.085 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 538.394 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 441.941 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 427.811 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 556.406 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 526.843 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 569.964 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 544.242 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 579.888 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 552.194 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 570.639 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 539.988 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 577.328 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 550.457 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 227 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 236.914 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 217 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 229.279 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 257.592 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.369 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 251.428 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.814 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 202 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.717 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 202 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.173 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 259.338 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 248.624 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 276.039 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 267.139 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 268.385 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 259.953 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 194 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 276.906 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 194 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 267.899 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 185 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 191.865 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 185 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 192.940 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 246 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 201.324 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 246 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 207.815 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 176 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 211.630 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 176 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 211.194 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 184 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 217.068 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 184 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 209.590 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 180 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 215.403 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 180 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 214.776 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 219.586 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 219.993 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 215.176 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 214.525 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 178 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 219.478 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 178 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 218.865 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 259.508 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.299 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 265.840 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 264.617 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 238.380 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 238.551 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 264.369 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 264.371 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 276.182 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 277.184 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 284.611 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 283.427 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 274.751 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 280.771 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 290.338 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 290.616 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 534.780 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 502.975 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 535.653 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 509.089 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 416.771 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 394.961 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 525.418 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 500.357 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 541.948 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 517.746 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 556.254 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 525.692 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 543.210 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 517.263 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 553.876 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 529.800 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 201 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 239.195 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 231.484 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 245.930 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 207 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 237.639 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 204 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 237.991 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 204 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 231.052 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 244.852 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 236.286 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.189 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.841 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 264.735 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 254.389 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 257.184 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 248.351 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 264.825 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 255.375 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 468.930 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 465.496 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 476.175 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 482.976 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 461.953 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 462.593 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 464.028 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 464.144 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 60 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 479.779 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 60 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 477.497 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 68 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 488.379 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 68 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 487.885 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 56 bytes spill stores, 60 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 476.546 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 56 bytes spill stores, 60 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 475.871 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 60 bytes spill stores, 64 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 483.043 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 60 bytes spill stores, 64 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 483.460 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.263 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.654 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 248.729 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.941 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.143 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.092 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 248.311 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.667 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 257.780 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 257.679 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 266.174 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 265.164 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.362 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.199 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 265.725 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 264.555 ms -sage_attention-torch-ext> [8/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f32_fuse_v_scale_attn.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 28 bytes gmem, 224 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 454.116 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 8 bytes stack frame, 4 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 8 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 450.317 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 537.396 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 8 bytes spill stores, 8 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 536.104 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 536.949 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 537.628 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 541.424 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 4 bytes spill stores, 4 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 541.008 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 468.252 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 40 bytes stack frame, 40 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 40 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 471.225 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 573.867 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 44 bytes spill stores, 36 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 574.696 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 469.310 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 466.100 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 567.081 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 569.499 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 236.441 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 239 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 233.116 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.673 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.201 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 237.993 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 241 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 232.422 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 239.934 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 231 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 235.001 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 243 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 251.274 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 243 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 251.161 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 256.662 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 244 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 257.186 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 250.344 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 240 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 251.082 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 255.857 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb1ELb0ELb0EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 233 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 255.752 ms -sage_attention-torch-ext> [12/12] Linking CXX shared module _sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> buildPhase completed in 3 minutes 42 seconds -sage_attention-torch-ext> Running phase: installPhase -sage_attention-torch-ext> install flags: -j21 install -sage_attention-torch-ext> [0/1] Install the project... -sage_attention-torch-ext> -- Install configuration: "Release" -sage_attention-torch-ext> -- Installing: /nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/_sage_attention_1369690_dirty/_sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> Running phase: fixupPhase -sage_attention-torch-ext> shrinking RPATHs of ELF executables and libraries in /nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext -sage_attention-torch-ext> shrinking /nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext/sage_attention/_sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> checking for references to /build/ in /nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext... -sage_attention-torch-ext> patching script interpreter paths in /nix/store/fh2zcbpbd5xsajp0dqgwmpw3hk9kfcdv-sage_attention-torch-ext -sage_attention-torch-ext> Running phase: installCheckPhase -sage_attention-torch-ext> no Makefile or custom installCheckPhase, doing nothing -sage_attention-torch-ext> Checking of ABI compatibility -sage_attention-torch-ext> 🐍 Checking for compatibility with manylinux_2_28 and Python ABI version 3.9 -sage_attention-torch-ext> ✅ No compatibility issues found -sage_attention-torch-ext> Checking loading kernel with get_kernel -sage_attention-torch-ext> Check whether the kernel can be loaded with get-kernel: sage_attention -sage_attention-torch-ext> [12/12] Linking CXX shared module _sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> buildPhase completed in 3 minutes 43 seconds -sage_attention-torch-ext> Running phase: installPhase -sage_attention-torch-ext> install flags: -j21 install -sage_attention-torch-ext> [0/1] Install the project... -sage_attention-torch-ext> -- Install configuration: "Release" -sage_attention-torch-ext> -- Installing: /nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/_sage_attention_1369690_dirty/_sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> Running phase: fixupPhase -sage_attention-torch-ext> shrinking RPATHs of ELF executables and libraries in /nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext -sage_attention-torch-ext> shrinking /nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext/sage_attention/_sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> checking for references to /build/ in /nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext... -sage_attention-torch-ext> patching script interpreter paths in /nix/store/8c8qa08gh6b21c551d4z6z6z1zp15jqr-sage_attention-torch-ext -sage_attention-torch-ext> Running phase: installCheckPhase -sage_attention-torch-ext> no Makefile or custom installCheckPhase, doing nothing -sage_attention-torch-ext> Checking of ABI compatibility -sage_attention-torch-ext> 🐍 Checking for compatibility with manylinux_2_28 and Python ABI version 3.9 -sage_attention-torch-ext> ✅ No compatibility issues found -sage_attention-torch-ext> Checking loading kernel with get_kernel -sage_attention-torch-ext> Check whether the kernel can be loaded with get-kernel: sage_attention -sage_attention-torch-ext> [9/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/sm89_qk_int8_sv_f8_accum_f16_attn_inst_buf.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 28 bytes gmem, 224 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 88 bytes stack frame, 156 bytes spill stores, 148 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 88 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 565.646 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 88 bytes stack frame, 156 bytes spill stores, 148 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 88 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 561.894 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 96 bytes spill stores, 92 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 472.572 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 80 bytes stack frame, 96 bytes spill stores, 92 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 80 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 473.206 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 32 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 537.302 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 32 bytes spill stores, 24 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 16 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 537.818 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 540.711 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 48 bytes stack frame, 32 bytes spill stores, 20 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 48 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 541.268 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 84 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 479.122 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 84 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 481.888 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 596.281 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 92 bytes spill stores, 88 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 595.258 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 164 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 496.615 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 176 bytes spill stores, 164 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 495.946 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 136 bytes spill stores, 128 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 596.177 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 136 bytes spill stores, 128 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 595.430 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 232.529 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 228.048 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 235.291 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 236.955 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 230.271 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 228.865 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 235.306 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode0ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 236.916 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 255.490 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 251.641 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.345 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.210 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 250.628 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb0ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 252.510 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.441 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z24qk_int_sv_f8_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit1EL8MaskMode1ELb1ELb0ELb0ELb1EEvPaS5_S5_PT9_PfS8_S8_S8_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 488 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 255.946 ms -sage_attention-torch-ext> [11/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/qk_int_sv_f16_cuda_sm80.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 11 bytes gmem, 88 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 564.271 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 528.211 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 567.301 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 537.719 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 440.514 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 426.754 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 553.227 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 526.362 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 569.300 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 544.159 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 581.426 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 551.856 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 570.959 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 542.792 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 580.170 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 552.088 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 227 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 237.978 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 217 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 230.451 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.427 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.941 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 251.539 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.539 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 202 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.921 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 202 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.075 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 259.891 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.931 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 278.386 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 267.722 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 267.410 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 259.995 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 194 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 277.844 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 194 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 268.692 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 185 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 192.172 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 185 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 191.798 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 246 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 202.309 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 246 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 208.954 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 176 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 210.573 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 176 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 212.166 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 184 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 217.046 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 184 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 211.601 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 180 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 215.271 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 180 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 215.576 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 220.112 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 219.610 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 214.027 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 214.428 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 178 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 219.749 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 178 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 219.153 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 260.099 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.387 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 268.371 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 265.606 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.062 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 239.485 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 267.143 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 267.449 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 278.266 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 277.525 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 285.609 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 286.443 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 278.234 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 286.253 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 294.132 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 294.033 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 540.389 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 509.537 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 540.716 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 514.407 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 422.445 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 400.817 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 531.424 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 505.847 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 549.355 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 521.168 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 557.945 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 532.220 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 546.641 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 519.650 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 558.367 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 530.083 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 201 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.074 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 232.210 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 245.509 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 207 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 238.357 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 204 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 237.668 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 204 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 230.888 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 243.525 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 237.689 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 255.042 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.985 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 262.963 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 254.173 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 254.833 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.697 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 262.415 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 254.606 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 470.452 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 467.379 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 478.690 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 481.126 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 461.883 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 463.891 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 462.930 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 462.210 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 60 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 477.580 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 60 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 479.471 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 68 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 488.864 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 68 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 489.588 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 56 bytes spill stores, 60 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 476.875 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 56 bytes spill stores, 60 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 478.730 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 60 bytes spill stores, 64 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 488.466 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 60 bytes spill stores, 64 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 486.969 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.338 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.900 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.556 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 246.948 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.532 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.661 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.878 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.969 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 259.093 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.858 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 267.126 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 266.400 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.847 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 258.737 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 265.989 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 266.591 ms -sage_attention-torch-ext> [12/12] Linking CXX shared module _sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> buildPhase completed in 3 minutes 51 seconds -sage_attention-torch-ext> Running phase: installPhase -sage_attention-torch-ext> install flags: -j21 install -sage_attention-torch-ext> [0/1] Install the project... -sage_attention-torch-ext> -- Install configuration: "Release" -sage_attention-torch-ext> -- Installing: /nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/_sage_attention_1369690_dirty/_sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> Running phase: fixupPhase -sage_attention-torch-ext> shrinking RPATHs of ELF executables and libraries in /nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext -sage_attention-torch-ext> shrinking /nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext/sage_attention/_sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> checking for references to /build/ in /nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext... -sage_attention-torch-ext> patching script interpreter paths in /nix/store/1p5vv3ybc6igpc3xin04i8p0d1hwbmw9-sage_attention-torch-ext -sage_attention-torch-ext> Running phase: installCheckPhase -sage_attention-torch-ext> no Makefile or custom installCheckPhase, doing nothing -sage_attention-torch-ext> Checking of ABI compatibility -sage_attention-torch-ext> 🐍 Checking for compatibility with manylinux_2_28 and Python ABI version 3.9 -sage_attention-torch-ext> ✅ No compatibility issues found -sage_attention-torch-ext> Checking loading kernel with get_kernel -sage_attention-torch-ext> Check whether the kernel can be loaded with get-kernel: sage_attention -sage_attention-torch-ext> [10/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/fused/fused.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 28 bytes gmem -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 32.522 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 29.255 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 25.043 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 23.716 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 4.458 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 4.239 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 4.108 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 4.084 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 4.022 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 3.975 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 3.378 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers -sage_attention-torch-ext> ptxas info : Compile time = 3.308 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.340 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.287 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.362 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.341 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.235 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.189 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.239 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.195 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.881 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 8.557 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 8.548 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 8.480 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.566 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 8.361 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 8.325 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 29 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 8.314 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 10.592 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.413 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 8.013 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.445 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 10.304 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.403 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.327 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.359 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 10.949 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.753 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.763 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.591 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 11.353 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.785 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.801 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_90a' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 28 registers, used 1 barriers, 132 bytes smem -sage_attention-torch-ext> ptxas info : Compile time = 7.682 ms -sage_attention-torch-ext> ptxas info : 28 bytes gmem, 224 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 22.378 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 20.630 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 16.377 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 16.501 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.329 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.129 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.032 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.049 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.007 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.940 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.533 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 20 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.533 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.934 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.860 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.864 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.846 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.896 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.923 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.959 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.899 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.645 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.823 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.817 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.811 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.561 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.901 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.871 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.872 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.902 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.913 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.901 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.867 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.345 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.864 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.906 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.868 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.702 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.975 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.009 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.478 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 32 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.901 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.079 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.991 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.004 ms -sage_attention-torch-ext> ptxas info : 28 bytes gmem, 224 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 39 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 29.801 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb1E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 39 registers, used 1 barriers, 392 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 27.683 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E13__nv_bfloat16EvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 21.867 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15MeanScaleKernelILj64ELb0E6__halfEvPT1_PaPfS4_fjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 260 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 21.890 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.132 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E13__nv_bfloat16EvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.062 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj128ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 21 registers, used 1 barriers, 32768 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.064 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z25TransposePadPermuteKernelILj64ELj64ELb1E6__halfEvPT2_S2_jjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 19 registers, used 1 barriers, 16384 bytes smem, 396 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 3.044 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.993 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E13__nv_bfloat16EvPT2_S2_P6__halfjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 16 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.959 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj128ELj64ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 18 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.543 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z13SubMeanKernelILj64ELj128ELj1E6__halfEvPT2_S2_PS0_jjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 18 registers, used 0 barriers, 412 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 2.561 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.849 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.823 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.873 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.815 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.938 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj32ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.858 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.917 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj16ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.860 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 38 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.552 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.813 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 25 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.812 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.778 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 40 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 10.568 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.866 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.927 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb1E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 8.427 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.414 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.852 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.880 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.865 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.326 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.880 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.919 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb0ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 6.902 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.837 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.024 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.033 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E13__nv_bfloat16EvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.505 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj128ELj2ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 33 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 9.974 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj128ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.021 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj128ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 27 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.030 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj' for 'sm_89' -sage_attention-torch-ext> ptxas info : Function properties for _Z15QuantInt8KernelILj64ELj64ELj1ELb1ELb0E6__halfEvPT4_S2_PaPffjjjjjjjjjjj -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 26 registers, used 1 barriers, 132 bytes smem, 432 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 7.007 ms -sage_attention-torch-ext> [11/12] Building CUDA object CMakeFiles/_sage_attention_1369690_dirty.dir/sage_attention/qattn/qk_int_sv_f16_cuda_sm80.cu.o -sage_attention-torch-ext> nvcc warning : incompatible redefinition for option 'threads', the last value of this option was used -sage_attention-torch-ext> ptxas info : 28 bytes gmem, 224 bytes cmem[4] -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 552.683 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 248 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 523.194 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 549.323 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 525.347 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 466.332 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 448.298 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 470.979 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 446.092 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 482.331 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 461.838 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 565.845 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 545.442 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 487.579 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 461.041 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 492.527 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 467.901 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 227 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 234.038 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 217 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 222.115 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.307 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.711 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.804 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 240.849 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 202 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 255.302 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 202 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 245.459 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 256.108 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 247.772 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 275.405 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 266.197 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 267.761 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 252.322 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 194 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 274.518 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb1EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 194 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 265.520 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 185 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 198.145 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 185 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 198.634 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 246 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 208.061 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 246 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 209.647 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 176 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 210.576 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 176 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 210.175 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 184 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 209.174 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 184 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 208.581 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 180 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 214.345 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 180 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 215.071 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 219.668 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 219.540 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 213.510 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 177 registers, used 1 barriers, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 213.305 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 178 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 217.553 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj16ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 16 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 178 registers, used 1 barriers, 16 bytes cumulative stack size, 480 bytes cmem[0] -sage_attention-torch-ext> ptxas info : Compile time = 218.382 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 255.931 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 255.049 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 262.130 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 249 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 262.696 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 237.992 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 237.915 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 262.410 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 262.170 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 275.393 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 274.460 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 281.105 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 281.974 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 275.665 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 273.540 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 280.176 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb1E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 280.231 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 446.182 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 253 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 426.698 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 529.447 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 431.382 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 443.656 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 423.599 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 508.858 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 482.621 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 457.277 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 435.240 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 465.834 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 254 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 443.837 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 455.826 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 250 registers, used 1 barriers, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 434.695 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 466.857 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 252 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 444.030 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 201 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 234.948 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 224.899 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 205 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 238.904 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 207 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 230.283 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 204 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 231.802 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 204 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 223.749 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 238.670 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 230.398 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.007 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.195 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 256.917 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 199 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 248.321 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 249.375 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 198 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.410 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS6_PS2_PT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 257.990 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2E6__halfLb0ES2_L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 200 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 248.705 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 455.769 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 454.613 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 470.853 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 32 bytes spill stores, 40 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 471.336 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 457.470 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 458.146 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 461.814 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 28 bytes spill stores, 32 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 459.742 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 60 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 476.324 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 64 bytes stack frame, 60 bytes spill stores, 68 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 64 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 475.340 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 68 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 484.900 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 112 bytes stack frame, 68 bytes spill stores, 76 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 112 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 485.926 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 56 bytes spill stores, 60 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 474.887 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 56 bytes stack frame, 56 bytes spill stores, 60 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 56 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 474.075 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 60 bytes spill stores, 64 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 483.415 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj128EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 96 bytes stack frame, 60 bytes spill stores, 64 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 96 bytes cumulative stack size, 480 bytes cmem[0], 16 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 482.843 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 238.801 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 236.384 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.917 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.693 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 235.138 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 235.253 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 241.941 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode0ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 242.014 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 254.167 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 253.268 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 260.712 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity3ELS1_3EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 259.851 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 253.822 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb0ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 253.470 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E13__nv_bfloat16L11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_P6__halfPT9_PfSA_SA_S9_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 260.374 ms -sage_attention-torch-ext> ptxas info : Compiling entry function '_Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf' for 'sm_80' -sage_attention-torch-ext> ptxas info : Function properties for _Z25qk_int_sv_f16_attn_kernelILj128ELj64ELj32ELj64ELj64EL8DataType1EL16QuantGranularity2ELS1_2EfLb0E6__halfL11ComputeUnit0EL8MaskMode1ELb1ELb0EEvPaS5_PS2_PT9_PfS9_S9_S8_jjjjjjjjjjjjjjjf -sage_attention-torch-ext> 32 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads -sage_attention-torch-ext> ptxas info : Used 255 registers, used 1 barriers, 32 bytes cumulative stack size, 480 bytes cmem[0], 8 bytes cmem[2] -sage_attention-torch-ext> ptxas info : Compile time = 259.803 ms -sage_attention-torch-ext> [12/12] Linking CXX shared module _sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> buildPhase completed in 5 minutes 8 seconds -sage_attention-torch-ext> Running phase: installPhase -sage_attention-torch-ext> install flags: -j21 install -sage_attention-torch-ext> [0/1] Install the project... -sage_attention-torch-ext> -- Install configuration: "Release" -sage_attention-torch-ext> -- Installing: /nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/_sage_attention_1369690_dirty/_sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> Running phase: fixupPhase -sage_attention-torch-ext> shrinking RPATHs of ELF executables and libraries in /nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext -sage_attention-torch-ext> shrinking /nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext/sage_attention/_sage_attention_1369690_dirty.abi3.so -sage_attention-torch-ext> checking for references to /build/ in /nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext... -sage_attention-torch-ext> patching script interpreter paths in /nix/store/n0xpmbwrclkfxs3sc50hr91s0180aqim-sage_attention-torch-ext -sage_attention-torch-ext> Running phase: installCheckPhase -sage_attention-torch-ext> no Makefile or custom installCheckPhase, doing nothing -sage_attention-torch-ext> Checking of ABI compatibility -sage_attention-torch-ext> 🐍 Checking for compatibility with manylinux_2_28 and Python ABI version 3.9 -sage_attention-torch-ext> ✅ No compatibility issues found -sage_attention-torch-ext> Checking loading kernel with get_kernel -sage_attention-torch-ext> Check whether the kernel can be loaded with get-kernel: sage_attention -building '/nix/store/ll29z68nd2l37pyv7ihc5sw3qv1wffjp-torch-ext-bundle.drv'... -building '/nix/store/dyhilzcxbx6rq6fhn6kpnhig1jf5rv4c-build-and-copy.drv'... diff --git a/pyproject.toml b/pyproject.toml deleted file mode 100644 index b9da95d03362688455d5ba0560560268b3ad6832..0000000000000000000000000000000000000000 --- a/pyproject.toml +++ /dev/null @@ -1,10 +0,0 @@ -[build-system] -requires = [ - "cmake>=3.26", - "ninja", - "packaging", - "setuptools>=61", - "torch", - "wheel", -] -build-backend = "setuptools.build_meta" \ No newline at end of file diff --git a/result b/result new file mode 120000 index 0000000000000000000000000000000000000000..b9f91cd7ac03337b497a652f01d67861e3c96495 --- /dev/null +++ b/result @@ -0,0 +1 @@ +/nix/store/874w0lpig5n3njbcbglp9pyj18wz5qzv-build-and-copy \ No newline at end of file diff --git a/setup.py b/setup.py deleted file mode 100644 index deae78d6eaad8f86977341074a9000cccd7e3ee4..0000000000000000000000000000000000000000 --- a/setup.py +++ /dev/null @@ -1,138 +0,0 @@ -import logging -import os -from shutil import which, move -import subprocess -import sys -from pathlib import Path - -from setuptools import Extension, find_packages, setup -from setuptools.command.build_ext import build_ext - -logger = logging.getLogger(__name__) - - -def is_sccache_available() -> bool: - return which("sccache") is not None - - -def is_ccache_available() -> bool: - return which("ccache") is not None - - -def is_ninja_available() -> bool: - return which("ninja") is not None - - -class CMakeExtension(Extension): - def __init__(self, name: str, sourcedir: str = "") -> None: - super().__init__(name, sources=[], py_limited_api=True) - self.sourcedir = os.fspath(Path(sourcedir).resolve()) - - -class CMakeBuild(build_ext): - def build_extension(self, ext: CMakeExtension) -> None: - ext_fullpath = Path.cwd() / self.get_ext_fullpath(ext.name) - extdir = ext_fullpath.parent.resolve() - - debug = int(os.environ.get("DEBUG", 0)) if self.debug is None else self.debug - cfg = "Debug" if debug else "Release" - - cmake_generator = os.environ.get("CMAKE_GENERATOR", "") - - # Set Python_EXECUTABLE instead if you use PYBIND11_FINDPYTHON - # EXAMPLE_VERSION_INFO shows you how to pass a value into the C++ code - # from Python. - cmake_args = [ - f"-DCMAKE_LIBRARY_OUTPUT_DIRECTORY={extdir}{os.sep}", - f"-DPython_EXECUTABLE={sys.executable}", - f"-DCMAKE_BUILD_TYPE={cfg}", # not used on MSVC, but no harm - ] - build_args = [] - if "CMAKE_ARGS" in os.environ: - cmake_args += [item for item in os.environ["CMAKE_ARGS"].split(" ") if item] - - if not cmake_generator or cmake_generator == "Ninja": - try: - import ninja - - ninja_executable_path = Path(ninja.BIN_DIR) / "ninja" - cmake_args += [ - "-GNinja", - f"-DCMAKE_MAKE_PROGRAM:FILEPATH={ninja_executable_path}", - ] - except ImportError: - pass - - if is_sccache_available(): - cmake_args += [ - "-DCMAKE_C_COMPILER_LAUNCHER=sccache", - "-DCMAKE_CXX_COMPILER_LAUNCHER=sccache", - "-DCMAKE_CUDA_COMPILER_LAUNCHER=sccache", - "-DCMAKE_HIP_COMPILER_LAUNCHER=sccache", - ] - elif is_ccache_available(): - cmake_args += [ - "-DCMAKE_C_COMPILER_LAUNCHER=ccache", - "-DCMAKE_CXX_COMPILER_LAUNCHER=ccache", - "-DCMAKE_CUDA_COMPILER_LAUNCHER=ccache", - "-DCMAKE_HIP_COMPILER_LAUNCHER=ccache", - ] - - num_jobs = os.getenv("MAX_JOBS", None) - if num_jobs is not None: - num_jobs = int(num_jobs) - logger.info("Using MAX_JOBS=%d as the number of jobs.", num_jobs) - else: - try: - # os.sched_getaffinity() isn't universally available, so fall - # back to os.cpu_count() if we get an error here. - num_jobs = len(os.sched_getaffinity(0)) - except AttributeError: - num_jobs = os.cpu_count() - - nvcc_threads = os.getenv("NVCC_THREADS", None) - if nvcc_threads is not None: - nvcc_threads = int(nvcc_threads) - logger.info( - "Using NVCC_THREADS=%d as the number of nvcc threads.", nvcc_threads - ) - else: - nvcc_threads = 1 - num_jobs = max(1, num_jobs // nvcc_threads) - - build_args += [f"-j{num_jobs}"] - if sys.platform == "win32": - build_args += ["--config", cfg] - - if nvcc_threads: - cmake_args += ["-DNVCC_THREADS={}".format(nvcc_threads)] - - build_temp = Path(self.build_temp) / ext.name - if not build_temp.exists(): - build_temp.mkdir(parents=True) - - subprocess.run( - ["cmake", ext.sourcedir, *cmake_args], cwd=build_temp, check=True - ) - subprocess.run( - ["cmake", "--build", ".", *build_args], cwd=build_temp, check=True - ) - if sys.platform == "win32": - # Move the dylib one folder up for discovery. - for filename in os.listdir(extdir / cfg): - move(extdir / cfg / filename, extdir / filename) - - - -setup( - name="sage_attention", - # The version is just a stub, it's not used by the final build artefact. - version="0.1.0", - ext_modules=[CMakeExtension("sage_attention._sage_attention_57cb7ec_dirty")], - cmdclass={"build_ext": CMakeBuild}, - packages=find_packages(where="torch-ext", include=["sage_attention*"]), - package_dir={"": "torch-ext"}, - zip_safe=False, - install_requires=["torch"], - python_requires=">=3.9", -) \ No newline at end of file diff --git a/torch-ext/sage_attention/_ops.py b/torch-ext/sage_attention/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..5f7c664eedba6a830cb5766aeb48420e89aaf90d --- /dev/null +++ b/torch-ext/sage_attention/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _sage_attention_bb220cf_dirty +ops = torch.ops._sage_attention_bb220cf_dirty + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_sage_attention_bb220cf_dirty::{op_name}" \ No newline at end of file diff --git a/torch-ext/sage_attention/_sage_attention_bb220cf_dirty.abi3.so b/torch-ext/sage_attention/_sage_attention_bb220cf_dirty.abi3.so new file mode 100755 index 0000000000000000000000000000000000000000..6ebc1477cced8dc26bb7e744a349a2f96d244444 --- /dev/null +++ b/torch-ext/sage_attention/_sage_attention_bb220cf_dirty.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f4812f32c4fbacb817bf7f4b7e10333e2fbac485dd9a27a2f0f2244e23f9fa1 +size 26051008