leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

v4.2 release. (#2587)

Browse Source 
* Fix default cluster callback values to 1 to avoid profiler failure when these values are not set in command line.

* v4.2 release.
This commit is contained in:
Junkai-Wu
2025-08-23 06:11:24 +08:00
committed by GitHub
parent 11cad1f67b
commit a49a78ffef
351 changed files with 28182 additions and 2032 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
python/cutlass_library/gemm_operation.py
View File

@@ -341,7 +341,7 @@ class GemmOperation:
Get the tile shape passed to the collective builder.
On Blackwell, this is different than the operation.tile_description.tile_shape.
"""
is_sm100_kernel = (self.arch == 100)
is_sm100_kernel = (self.arch == 100 or self.arch == 103)
if not is_sm100_kernel:
return self.tile_description.tile_shape
@@ -995,6 +995,24 @@ ${compile_guard_end}
epi_tile_mn = "cute::Shape<cute::_128,cute::_64>"
if not is_no_smem_epilogue:
epilogue_schedule_type = EpilogueScheduleTag[to_grouped_schedule(EpilogueScheduleType.TmaWarpSpecialized2Sm, grouped)]
if cta_n == 256 and operation.kernel_schedule == KernelScheduleType.BlockScaledMxNvf4UltraTmaWarpSpecialized1SmVs32Sm103:
epi_tile_mn = "cute::Shape<cute::_128,cute::_64>"
if not is_no_smem_epilogue:
epilogue_schedule_type = EpilogueScheduleTag[EpilogueScheduleType.TmaWarpSpecialized1Sm]
if cta_n == 256 and operation.kernel_schedule == KernelScheduleType.BlockScaledMxNvf4UltraTmaWarpSpecialized2SmVs32Sm103:
epi_tile_mn = "cute::Shape<cute::_128,cute::_64>"
if not is_no_smem_epilogue:
epilogue_schedule_type = EpilogueScheduleTag[EpilogueScheduleType.TmaWarpSpecialized2Sm]
if cta_n == 256 and operation.kernel_schedule == KernelScheduleType.BlockScaledMxNvf4UltraTmaWarpSpecialized1SmVs16Sm103:
epi_tile_mn = "cute::Shape<cute::_128,cute::_64>"
if not is_no_smem_epilogue:
epilogue_schedule_type = EpilogueScheduleTag[EpilogueScheduleType.TmaWarpSpecialized1Sm]
if cta_n == 256 and operation.kernel_schedule == KernelScheduleType.BlockScaledMxNvf4UltraTmaWarpSpecialized2SmVs16Sm103:
epi_tile_mn = "cute::Shape<cute::_128,cute::_64>"
if not is_no_smem_epilogue:
epilogue_schedule_type = EpilogueScheduleTag[EpilogueScheduleType.TmaWarpSpecialized2Sm]
element_a = f'cute::tuple<{str(element_a)},{str(DataTypeTag[operation.ScaleFactorA])}>'
element_b = f'cute::tuple<{str(element_b)},{str(DataTypeTag[operation.ScaleFactorB])}>'