leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 27 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
2025-08-22 18:11:24 -04:00
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
test/python/cutlass/gemm/gemm_f64_sm90.py
+9 -9
View File
@@ -38,19 +38,19 @@ from functools import partial
import logging
import unittest
import cutlass
from cutlass.backend.utils.device import device_cc
import cutlass_cppgen
from cutlass_cppgen.backend.utils.device import device_cc
from utils import LayoutCombination, add_test_gemm
cutlass.set_log_level(logging.WARNING)
cutlass_cppgen.set_log_level(logging.WARNING)
cc = 90
dtype = cutlass.DataType.f64
dtype = cutlass_cppgen.DataType.f64
@unittest.skipIf(device_cc() < cc, 'Device compute capability is insufficient for SM90 tests.')
@unittest.skipIf(cutlass.utils.datatypes.torch_type(dtype) is None, f'Version of torch installed does not contain a datatype match for {dtype}')
@unittest.skipIf(cutlass_cppgen.utils.datatypes.torch_type(dtype) is None, f'Version of torch installed does not contain a datatype match for {dtype}')
class GemmF64Sm90(unittest.TestCase):
"""
Wrapper class to which tests will be added dynamically in __main__
@@ -61,10 +61,10 @@ class GemmF64Sm90(unittest.TestCase):
add_test_specialized = partial(add_test_gemm, cls=GemmF64Sm90, alignments=[1, 1, 1], cluster_shape=[1, 1, 1],
element=dtype, element_output=dtype, element_accumulator=dtype, compilation_modes=['nvcc'])
add_test_specialized(opclass=cutlass.OpcodeClass.TensorOp, layouts=LayoutCombination.NNT, threadblock_shape=[128, 128, 32], stages=3)
add_test_specialized(opclass=cutlass.OpcodeClass.TensorOp, layouts=LayoutCombination.TNN, threadblock_shape=[128, 128, 32], stages=3)
add_test_specialized( opclass=cutlass.OpcodeClass.Simt, layouts=LayoutCombination.NNN, threadblock_shape=[128, 128, 8], stages=2)
add_test_specialized( opclass=cutlass.OpcodeClass.Simt, layouts=LayoutCombination.TTT, threadblock_shape=[ 64, 128, 8], stages=2)
add_test_specialized(opclass=cutlass_cppgen.OpcodeClass.TensorOp, layouts=LayoutCombination.NNT, threadblock_shape=[128, 128, 32], stages=3)
add_test_specialized(opclass=cutlass_cppgen.OpcodeClass.TensorOp, layouts=LayoutCombination.TNN, threadblock_shape=[128, 128, 32], stages=3)
add_test_specialized( opclass=cutlass_cppgen.OpcodeClass.Simt, layouts=LayoutCombination.NNN, threadblock_shape=[128, 128, 8], stages=2)
add_test_specialized( opclass=cutlass_cppgen.OpcodeClass.Simt, layouts=LayoutCombination.TTT, threadblock_shape=[ 64, 128, 8], stages=2)
if __name__ == '__main__':