leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 15 Packages Projects Releases Wiki Activity

CUTLASS 2.10 updates (#622)

Browse Source 
Co-authored-by: Aniket Shivam <ashivam@nvidia.com>
This commit is contained in:
ANIKET SHIVAM
2022-09-12 21:26:30 -04:00
committed by GitHub
co-authored by Aniket Shivam
parent beae168f90
commit e773429f7e
96 changed files with 8365 additions and 1667 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.py
+28 -16
View File
@@ -30,15 +30,18 @@ class Conv2dDgradImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float16)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -68,15 +71,18 @@ class Conv2dDgradImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float16)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -106,15 +112,18 @@ class Conv2dDgradImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float16)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -156,15 +165,18 @@ class Conv2dDgradImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float16)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py
+28 -16
View File
@@ -29,15 +29,18 @@ class Conv2dDgradImplicitGemmF16nhwcF16nhwcF32nhwcTensorOpF32SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 32], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -67,15 +70,18 @@ class Conv2dDgradImplicitGemmF16nhwcF16nhwcF32nhwcTensorOpF32SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 32], stages=4,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -105,15 +111,18 @@ class Conv2dDgradImplicitGemmF16nhwcF16nhwcF32nhwcTensorOpF32SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -143,15 +152,18 @@ class Conv2dDgradImplicitGemmF16nhwcF16nhwcF32nhwcTensorOpF32SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=4,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_f32nhwc_f32nhwc_f32nhwc_simt_f32_sm80.py
+14 -8
View File
@@ -30,15 +30,18 @@ class Conv2dDgradImplicitGemmF32nhwcF32nhwcF32nhwcSimtF32SM80(unittest.TestCase)
tile_description = TileDescription(
threadblock_shape=[128, 128, 8], stages=4,
warp_count=[4, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -68,15 +71,18 @@ class Conv2dDgradImplicitGemmF32nhwcF32nhwcF32nhwcSimtF32SM80(unittest.TestCase)
tile_description = TileDescription(
threadblock_shape=[128, 128, 8], stages=4,
warp_count=[2, 4, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_sm80.py
+14 -8
View File
@@ -29,15 +29,18 @@ class Conv2dDgradImplicitGemmTF32nhwcTF32nhwcTF32nhwcTensorOpF32SM80(unittest.Te
tile_description = TileDescription(
threadblock_shape=[128, 128, 16], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -67,15 +70,18 @@ class Conv2dDgradImplicitGemmTF32nhwcTF32nhwcTF32nhwcTensorOpF32SM80(unittest.Te
tile_description = TileDescription(
threadblock_shape=[128, 128, 16], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Unity,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Unity,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_fprop_few_channels_f16nhwc_f16nhwc_f16nhwc_tensor_op_f32_sm80.py
+14 -8
View File
@@ -97,15 +97,18 @@ class Conv2dFpropFewChannelsF16NHWCF16NHWCF16HNWCTensorOpF32SM80(unittest.TestCa
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.few_channels,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -135,15 +138,18 @@ class Conv2dFpropFewChannelsF16NHWCF16NHWCF16HNWCTensorOpF32SM80(unittest.TestCa
tile_description = TileDescription(
threadblock_shape=[128, 128, 32], stages=2,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.few_channels,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_fprop_fixed_channels_f16nhwc_f16nhwc_f16nhwc_tensor_op_f32_sm80.py
+21 -12
View File
@@ -79,15 +79,18 @@ class Conv2dFpropFixedChannelsF16NHWCF16NHWCF16HNWCTensorOpF32SM80(unittest.Test
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.fixed_channels,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -117,15 +120,18 @@ class Conv2dFpropFixedChannelsF16NHWCF16NHWCF16HNWCTensorOpF32SM80(unittest.Test
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.fixed_channels,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -155,15 +161,18 @@ class Conv2dFpropFixedChannelsF16NHWCF16NHWCF16HNWCTensorOpF32SM80(unittest.Test
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.fixed_channels,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.py
+35 -20
View File
@@ -29,15 +29,18 @@ class Conv2dFpropImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float16)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -67,15 +70,18 @@ class Conv2dFpropImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float16)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -105,15 +111,18 @@ class Conv2dFpropImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float16)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -173,15 +182,18 @@ class Conv2dFpropImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float16)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -241,15 +253,18 @@ class Conv2dFpropImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float16)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py
+7 -4
View File
@@ -29,15 +29,18 @@ class Conv2dFpropImplicitGemmF16nhwcF16nhwcF32nhwcTensorOpF32SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_f32nhwc_f32nhwc_f32nhwc_simt_f32_sm80.py
+14 -8
View File
@@ -30,15 +30,18 @@ class Conv2dFpropImplicitGemmF32nhwcF32nhwcF32nhwcSimtF32SM80(unittest.TestCase)
tile_description = TileDescription(
threadblock_shape=[128, 128, 8], stages=4,
warp_count=[4, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle2
)
@@ -68,15 +71,18 @@ class Conv2dFpropImplicitGemmF32nhwcF32nhwcF32nhwcSimtF32SM80(unittest.TestCase)
tile_description = TileDescription(
threadblock_shape=[128, 128, 8], stages=4,
warp_count=[2, 4, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_sm80.py
+14 -8
View File
@@ -29,15 +29,18 @@ class Conv2dFpropImplicitGemmTF32nhwcTF32nhwcTF32nhwcTensorOpF32SM80(unittest.Te
tile_description = TileDescription(
threadblock_shape=[128, 128, 16], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -67,15 +70,18 @@ class Conv2dFpropImplicitGemmTF32nhwcTF32nhwcTF32nhwcTensorOpF32SM80(unittest.Te
tile_description = TileDescription(
threadblock_shape=[128, 128, 16], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.fprop, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_strided_dgrad_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py
+35 -20
View File
@@ -29,15 +29,18 @@ class Conv2dStridedDgradImplicitGemmF16NHWCF16NHWCF32NHWCTensorOpF32SM80(unittes
tile_description = TileDescription(
threadblock_shape=[128, 128, 32], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.StridedDgradIdentitySwizzle1
)
@@ -67,15 +70,18 @@ class Conv2dStridedDgradImplicitGemmF16NHWCF16NHWCF32NHWCTensorOpF32SM80(unittes
tile_description = TileDescription(
threadblock_shape=[128, 256, 64], stages=3,
warp_count=[2, 4, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.StridedDgradIdentitySwizzle1
)
@@ -105,15 +111,18 @@ class Conv2dStridedDgradImplicitGemmF16NHWCF16NHWCF32NHWCTensorOpF32SM80(unittes
tile_description = TileDescription(
threadblock_shape=[128, 128, 32], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.StridedDgradIdentitySwizzle1
)
@@ -155,15 +164,18 @@ class Conv2dStridedDgradImplicitGemmF16NHWCF16NHWCF32NHWCTensorOpF32SM80(unittes
tile_description = TileDescription(
threadblock_shape=[128, 128, 32], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.StridedDgradIdentitySwizzle1
)
@@ -193,15 +205,18 @@ class Conv2dStridedDgradImplicitGemmF16NHWCF16NHWCF32NHWCTensorOpF32SM80(unittes
tile_description = TileDescription(
threadblock_shape=[128, 128, 32], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.dgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.StridedDgradIdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.py
+16 -8
View File
@@ -29,15 +29,19 @@ class Conv2dWgradImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment, math_inst.element_accumulator,
cutlass.float16
)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -67,15 +71,19 @@ class Conv2dWgradImplicitGemmF16nhwcF16nhwcF16nhwcTensorOpF16SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 64], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment, math_inst.element_accumulator,
cutlass.float16
)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float16, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py
+35 -20
View File
@@ -29,15 +29,18 @@ class Conv2dWgradImplicitGemmF16nhwcF16nhwcF32nhwcTensorOpF32SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 16], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -67,15 +70,18 @@ class Conv2dWgradImplicitGemmF16nhwcF16nhwcF32nhwcTensorOpF32SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 16], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -105,15 +111,18 @@ class Conv2dWgradImplicitGemmF16nhwcF16nhwcF32nhwcTensorOpF32SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[64, 256, 32], stages=3,
warp_count=[1, 4, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -143,15 +152,18 @@ class Conv2dWgradImplicitGemmF16nhwcF16nhwcF32nhwcTensorOpF32SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 16], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -193,15 +205,18 @@ class Conv2dWgradImplicitGemmF16nhwcF16nhwcF32nhwcTensorOpF32SM80(unittest.TestC
tile_description = TileDescription(
threadblock_shape=[128, 128, 16], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_f32nhwc_f32nhwc_f32nhwc_simt_f32_sm80.py
+14 -8
View File
@@ -30,15 +30,18 @@ class Conv2dWgradImplicitGemmF32nhwcF32nhwcF32nhwcSimtF32SM80(unittest.TestCase)
tile_description = TileDescription(
threadblock_shape=[128, 128, 8], stages=4,
warp_count=[2, 4, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.analytic,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -68,15 +71,18 @@ class Conv2dWgradImplicitGemmF32nhwcF32nhwcF32nhwcSimtF32SM80(unittest.TestCase)
tile_description = TileDescription(
threadblock_shape=[128, 128, 8], stages=4,
warp_count=[2, 4, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_sm80.py
+14 -8
View File
@@ -29,15 +29,18 @@ class Conv2dWgradImplicitGemmTF32nhwcTF32nhwcTF32nhwcTensorOpF32SM80(unittest.Te
tile_description = TileDescription(
threadblock_shape=[128, 128, 16], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -67,15 +70,18 @@ class Conv2dWgradImplicitGemmTF32nhwcTF32nhwcTF32nhwcTensorOpF32SM80(unittest.Te
tile_description = TileDescription(
threadblock_shape=[128, 128, 32], stages=3,
warp_count=[2, 2, 1],
math_instruction=math_inst,
min_compute=80, max_compute=80
math_instruction=math_inst
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
operation = Conv2dOperation(
conv_kind=cutlass.conv.Operator.wgrad, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=80, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=cutlass.float32, stride_support=StrideSupport.Strided,
epilogue_functor=EpilogueFunctor.LinearCombination,
stride_support=StrideSupport.Strided,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
tools/library/scripts/pycutlass/test/example/run_all_example.sh
+80
View File
@@ -0,0 +1,80 @@
pushd $CUTLASS_PATH/examples/40_cutlass_py/
python gemm.py -i 8 8 4 -ta float64 -tb float64 -tc float64 -tacc float64 -m multiply_add -op TensorOp -b 32 32 16 -s 4 -w 2 2 1 -cc 80 -la ColumnMajor -aa 1 -lb RowMajor -ab 1 -lc RowMajor -ac 1 -te float64 -ep LinearCombination -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 1
python gemm.py -i 8 8 4 -ta float64 -tb float64 -tc float64 -tacc float64 -m multiply_add -op TensorOp -b 64 64 16 -s 4 -w 2 2 1 -cc 80 -la RowMajor -aa 1 -lb ColumnMajor -ab 1 -lc RowMajor -ac 1 -te float64 -ep LinearCombination -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 2
python gemm.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add_fast_bf16 -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la RowMajor -aa 4 -lb ColumnMajor -ab 4 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 1
python gemm.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 4 -lb ColumnMajor -ab 4 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm GemmSplitKParallel -k 2
python gemm.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add_fast_f32 -op TensorOp -b 64 64 32 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 4 -lb ColumnMajor -ab 4 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 4
python gemm.py -i 16 8 16 -ta float16 -tb float16 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 8 -lb RowMajor -ab 8 -lc ColumnMajor -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle4 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 1
python gemm.py -i 16 8 16 -ta float16 -tb float16 -tc float16 -tacc float32 -m multiply_add -op TensorOp -b 128 128 64 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 8 -lb ColumnMajor -ab 8 -lc RowMajor -ac 8 -te float32 -ep LinearCombination -sw IdentitySwizzle2 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 2
python gemm.py -i 16 8 16 -ta float16 -tb float16 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 256 128 64 -s 3 -w 4 2 1 -cc 80 -la ColumnMajor -aa 8 -lb ColumnMajor -ab 8 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm GemmSplitKParallel -k 3
python gemm.py -i 16 8 16 -ta bfloat16 -tb bfloat16 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 64 128 64 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 8 -lb ColumnMajor -ab 8 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle2 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm GemmSplitKParallel -k 5
python gemm.py -i 16 8 32 -ta int8 -tb int8 -tc int8 -tacc int32 -m multiply_add -op TensorOp -b 128 128 128 -s 3 -w 2 2 1 -cc 80 -la RowMajor -aa 16 -lb ColumnMajor -ab 16 -lc RowMajor -ac 16 -te float32 -ep FastLinearCombinationClamp -sw IdentitySwizzle2 -p 512 512 512 -alpha 1.0 -beta 0.0 -gm Gemm -k 1
python gemm_grouped.py -i 16 8 16 -ta float16 -tb float16 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 8 -lb ColumnMajor -ab 8 -lc ColumnMajor -ac 4 -te float32 -ep LinearCombination -p ./grouped_gemm_problem_size.csv -alpha 1.0 -beta 0.0 -pm Device
python gemm_grouped.py -i 8 8 4 -ta float64 -tb float64 -tc float64 -tacc float64 -m multiply_add -op TensorOp -b 64 64 16 -s 4 -w 2 2 1 -cc 80 -la RowMajor -aa 1 -lb RowMajor -ab 1 -lc ColumnMajor -ac 1 -te float64 -ep LinearCombination -p ./grouped_gemm_problem_size.csv -alpha 1.0 -beta 1.0 -pm Host
python gemm_grouped.py -i 1 1 1 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add -op Simt -b 128 64 8 -s 4 -w 2 2 1 -cc 80 -la RowMajor -aa 1 -lb RowMajor -ab 1 -lc RowMajor -ac 1 -te float32 -ep LinearCombination -p ./grouped_gemm_problem_size.csv -alpha 2.0 -beta 1.0 -pm Device
python gemm_grouped.py -i 16 8 16 -ta float16 -tb float16 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 8 -lb ColumnMajor -ab 8 -lc ColumnMajor -ac 4 -te float32 -ep LinearCombination -p ./grouped_gemm_problem_size.csv -alpha 2.0 -beta 1.0 -pm Device
python conv2d.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 16 -s 3 -w 2 2 1 -cc 80 -la TensorNHWC -aa 4 -lb TensorNHWC -ab 4 -lc TensorNHWC -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -co fprop -st Strided -ia optimized -sm Serial -k 1 -nhwc 1 13 17 8 -krsc 24 3 3 8 -pad 0 0 0 0 -stride 2 2 -dilation 1 1 -alpha 1.0 -beta 0.0
python conv2d.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 16 -s 3 -w 2 2 1 -cc 80 -la TensorNHWC -aa 2 -lb TensorNHWC -ab 2 -lc TensorNHWC -ac 2 -te float32 -ep LinearCombination -sw IdentitySwizzle2 -co fprop -st Strided -ia optimized -sm Serial -k 2 -nhwc 1 4 4 12 -krsc 8 3 3 12 -pad 0 0 0 0 -stride 3 3 -dilation 1 1 -alpha 1.0 -beta 1.0
python conv2d.py -i 1 1 1 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add -op Simt -b 128 128 8 -s 4 -w 4 2 1 -cc 80 -la TensorNHWC -aa 4 -lb TensorNHWC -ab 4 -lc TensorNHWC -ac 1 -te float32 -ep LinearCombination -sw IdentitySwizzle4 -co fprop -st Strided -ia analytic -sm Parallel -k 3 -nhwc 1 71 80 32 -krsc 64 5 5 32 -pad 2 2 2 2 -stride 2 2 -dilation 1 1 -alpha 1.0 -beta 1.0
python conv2d.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la TensorNHWC -aa 1 -lb TensorNHWC -ab 1 -lc TensorNHWC -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -co wgrad -st Strided -ia optimized -sm Serial -k 1 -nhwc 1 8 8 1 -krsc 1 3 3 1 -pad 1 1 1 1 -stride 1 1 -dilation 1 1 -alpha 1.0 -beta 0.0
python conv2d.py -i 1 1 1 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add -op Simt -b 128 128 8 -s 4 -w 2 4 1 -cc 80 -la TensorNHWC -aa 4 -lb TensorNHWC -ab 4 -lc TensorNHWC -ac 1 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -co wgrad -st Strided -ia optimized -sm Serial -k 2 -nhwc 1 27 27 256 -krsc 512 3 3 256 -pad 1 1 1 1 -stride 2 1 -dilation 1 1 -alpha 1.0 -beta 0.0
python conv2d.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 16 -s 3 -w 2 2 1 -cc 80 -la TensorNHWC -aa 4 -lb TensorNHWC -ab 4 -lc TensorNHWC -ac 4 -te float32 -ep LinearCombination -sw StridedDgradIdentitySwizzle1 -co dgrad -st Strided -ia optimized -sm Serial -k 2 -nhwc 1 27 27 256 -krsc 512 3 3 256 -pad 1 1 1 1 -stride 2 1 -dilation 1 1 -alpha 1.0 -beta 0.0
python conv2d.py -i 16 8 16 -ta float16 -tb float16 -tc float16 -tacc float32 -m multiply_add -op TensorOp -b 128 128 64 -s 3 -w 2 2 1 -cc 80 -la TensorNHWC -aa 8 -lb TensorNHWC -ab 8 -lc TensorNHWC -ac 8 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -co fprop -st Strided -ia optimized -sm Serial -k 1 -nhwc 1 27 27 256 -krsc 512 3 3 256 -pad 1 1 1 1 -stride 2 1 -dilation 1 1 -alpha 1.0 -beta 0.0
python conv2d.py -i 16 8 16 -ta float16 -tb float16 -tc float16 -tacc float32 -m multiply_add -op TensorOp -b 128 128 64 -s 3 -w 2 2 1 -cc 80 -la TensorNHWC -aa 2 -lb TensorNHWC -ab 2 -lc TensorNHWC -ac 8 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -co fprop -st Strided -ia few_channels -sm Serial -k 1 -nhwc 1 16 16 2 -krsc 16 3 3 2 -pad 1 1 1 1 -stride 2 2 -dilation 1 1 -alpha 1.0 -beta 0.0
python conv2d.py -i 16 8 16 -ta float16 -tb float16 -tc float16 -tacc float32 -m multiply_add -op TensorOp -b 128 128 64 -s 3 -w 2 2 1 -cc 80 -la TensorNHWC -aa 8 -lb TensorNHWC -ab 8 -lc TensorNHWC -ac 8 -te float32 -ep LinearCombination -sw IdentitySwizzle2 -co fprop -st Strided -ia fixed_channels -sm Serial -k 1 -nhwc 1 8 8 8 -krsc 16 3 3 8 -pad 1 1 1 1 -stride 2 2 -dilation 1 1 -alpha 1.0 -beta 0.0
python conv2d.py -i 16 8 16 -ta float16 -tb float16 -tc float16 -tacc float32 -m multiply_add -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la TensorNHWC -aa 4 -lb TensorNHWC -ab 4 -lc TensorNHWC -ac 4 -te float32 -ep LinearCombination -sw StridedDgradIdentitySwizzle1 -co dgrad -st Strided -ia optimized -sm Serial -k 1 -nhwc 1 56 56 12 -krsc 8 1 1 12 -pad 0 0 0 0 -stride 2 2 -dilation 1 1 -alpha 1.0 -beta 0.0
python gemm.py -i 8 8 4 -ta float64 -tb float64 -tc float64 -tacc float64 -m multiply_add -op TensorOp -b 32 32 16 -s 4 -w 2 2 1 -cc 80 -la ColumnMajor -aa 1 -lb RowMajor -ab 1 -lc RowMajor -ac 1 -te float64 -ep LinearCombination -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 1 -bias -activ relu
python gemm.py -i 8 8 4 -ta float64 -tb float64 -tc float64 -tacc float64 -m multiply_add -op TensorOp -b 64 64 16 -s 4 -w 2 2 1 -cc 80 -la RowMajor -aa 1 -lb ColumnMajor -ab 1 -lc RowMajor -ac 1 -te float64 -ep LinearCombination -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 2 -bias -activ leaky_relu -activ_arg 0.2
python gemm.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 4 -lb ColumnMajor -ab 4 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm GemmSplitKParallel -k 2 -bias -activ tanh
python gemm_grouped.py -i 8 8 4 -ta float64 -tb float64 -tc float64 -tacc float64 -m multiply_add -op TensorOp -b 64 64 16 -s 4 -w 2 2 1 -cc 80 -la RowMajor -aa 1 -lb RowMajor -ab 1 -lc ColumnMajor -ac 1 -te float64 -ep LinearCombination -p ./grouped_gemm_problem_size.csv -alpha 0.0 -beta 0.5 -pm Host -bias -activ sigmoid -bias -activ sigmoid
python conv2d.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 16 -s 3 -w 2 2 1 -cc 80 -la TensorNHWC -aa 2 -lb TensorNHWC -ab 2 -lc TensorNHWC -ac 2 -te float32 -ep LinearCombination -sw IdentitySwizzle2 -co fprop -st Strided -ia optimized -sm Serial -k 2 -nhwc 1 4 4 12 -krsc 8 3 3 12 -pad 0 0 0 0 -stride 3 3 -dilation 1 1 -alpha 0.0 -beta 0.5 -bias -activ silu
python conv2d.py -i 16 8 16 -ta float16 -tb float16 -tc float16 -tacc float32 -m multiply_add -op TensorOp -b 128 128 64 -s 3 -w 2 2 1 -cc 80 -la TensorNHWC -aa 2 -lb TensorNHWC -ab 2 -lc TensorNHWC -ac 8 -te float32 -ep LinearCombination -sw IdentitySwizzle1 -co fprop -st Strided -ia few_channels -sm Serial -k 1 -nhwc 1 16 16 2 -krsc 16 3 3 2 -pad 1 1 1 1 -stride 2 2 -dilation 1 1 -alpha 0.0 -beta 0.5 -bias -activ hardswish
python gemm.py -i 16 8 16 -ta bfloat16 -tb bfloat16 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 64 128 64 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 8 -lb ColumnMajor -ab 8 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle2 -p 512 256 128 -alpha 0.0 -beta 0.5 -gm GemmSplitKParallel -k 5 -bias -activ gelu
python gemm.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add_fast_bf16 -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la RowMajor -aa 4 -lb ColumnMajor -ab 4 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -sw BatchedIdentitySwizzle -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Batched -k 1 -batch 3
python gemm.py -i 16 8 16 -ta float16 -tb float16 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 8 -lb RowMajor -ab 8 -lc ColumnMajor -ac 4 -te float32 -ep LinearCombination -sw IdentitySwizzle4 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Array -k 1 -batch 2
python gemm.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add_fast_bf16 -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la RowMajor -aa 4 -lb ColumnMajor -ab 4 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -epv RowBroadcast -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 1
python gemm.py -i 8 8 4 -ta float64 -tb float64 -tc float64 -tacc float64 -m multiply_add -op TensorOp -b 32 32 16 -s 4 -w 2 2 1 -cc 80 -la ColumnMajor -aa 1 -lb RowMajor -ab 1 -lc RowMajor -ac 1 -te float64 -ep LinearCombination -epv ColumnBroadcast -sw IdentitySwizzle1 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 1
python gemm.py -i 16 8 16 -ta float16 -tb float16 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 8 -lb RowMajor -ab 8 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -epv RowReduction -sw IdentitySwizzle4 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 1
python gemm.py -i 16 8 16 -ta bfloat16 -tb bfloat16 -tc float32 -tacc float32 -m multiply_add -op TensorOp -b 64 128 64 -s 3 -w 2 2 1 -cc 80 -la ColumnMajor -aa 8 -lb ColumnMajor -ab 8 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -epv ColumnReduction -sw IdentitySwizzle2 -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Gemm -k 1
python gemm.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add_fast_bf16 -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la RowMajor -aa 4 -lb ColumnMajor -ab 4 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -epv RowReduction -sw BatchedIdentitySwizzle -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Batched -k 1 -batch 3
python gemm.py -i 16 8 8 -ta float32 -tb float32 -tc float32 -tacc float32 -m multiply_add_fast_bf16 -op TensorOp -b 128 128 32 -s 3 -w 2 2 1 -cc 80 -la RowMajor -aa 4 -lb ColumnMajor -ab 4 -lc RowMajor -ac 4 -te float32 -ep LinearCombination -epv ColumnBroadcast -sw BatchedIdentitySwizzle -p 512 256 128 -alpha 1.0 -beta 0.5 -gm Array -k 1 -batch 3
popd
tools/library/scripts/pycutlass/test/frontend/test_frontend.py
+9 -5
View File
@@ -49,7 +49,7 @@ class Test_Frontend(unittest.TestCase):
tile_description = TileDescription(
[128, 128, 8], 4, [2, 4, 1],
math_inst, 80, 80
math_inst
)
A = TensorDescription(
@@ -64,10 +64,14 @@ class Test_Frontend(unittest.TestCase):
cutlass.float32, cutlass.RowMajor, 1
)
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
self.operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=cutlass.float32,
epilogue_functor=EpilogueFunctor.LinearCombination,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor,
swizzling_functor=cutlass.IdentitySwizzle1
)
@@ -89,7 +93,7 @@ class Test_Frontend(unittest.TestCase):
arguments = GemmArguments(
operation=self.operation, problem_size=problem_size,
A=tensor_A, B=tensor_B, C=tensor_C, D=tensor_D,
output_op=LinearCombinationFunctorArguments(alpha, beta),
output_op=self.operation.epilogue_type(alpha, beta),
gemm_mode=cutlass.gemm.Mode.Gemm, split_k_splices=1
)
@@ -119,7 +123,7 @@ class Test_Frontend(unittest.TestCase):
arguments = GemmArguments(
operation=self.operation, problem_size=problem_size,
A=tensor_A, B=tensor_B, C=tensor_C, D=tensor_D,
output_op=LinearCombinationFunctorArguments(alpha, beta),
output_op=self.operation.epilogue_type(alpha, beta),
gemm_mode=cutlass.gemm.Mode.Gemm, split_k_splices=1
)
tools/library/scripts/pycutlass/test/gemm/gemm_bf16_sm80.py
+10 -10
View File
@@ -17,7 +17,7 @@ class GemmBF16TensorOpSm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[64, 128, 64],
stages=4, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -33,15 +33,15 @@ class GemmBF16TensorOpSm80(unittest.TestCase):
alignment=4
)
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -58,7 +58,7 @@ class GemmBF16TensorOpSm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[64, 128, 32],
stages=6, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -74,15 +74,15 @@ class GemmBF16TensorOpSm80(unittest.TestCase):
alignment=8
)
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, cutlass.float32)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
tools/library/scripts/pycutlass/test/gemm/gemm_f16_sm80.py
+50 -30
View File
@@ -18,7 +18,7 @@ class GemmF16Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 128, 32],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -36,13 +36,15 @@ class GemmF16Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.BatchedIdentitySwizzle
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor,
direct_store=True
)
@@ -60,7 +62,7 @@ class GemmF16Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 128, 64],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -78,13 +80,15 @@ class GemmF16Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -101,7 +105,7 @@ class GemmF16Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 256, 64],
stages=3, warp_count=[2, 4, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -119,13 +123,15 @@ class GemmF16Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -142,7 +148,7 @@ class GemmF16Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[256, 128, 64],
stages=3, warp_count=[4, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -160,13 +166,15 @@ class GemmF16Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -183,7 +191,7 @@ class GemmF16Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 64, 64],
stages=3, warp_count=[2, 1, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -201,13 +209,15 @@ class GemmF16Sm80(unittest.TestCase):
element_epilogue = cutlass.float16
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -224,7 +234,7 @@ class GemmF16Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[64, 64, 32],
stages=10, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -242,13 +252,15 @@ class GemmF16Sm80(unittest.TestCase):
element_epilogue = cutlass.float16
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -265,7 +277,7 @@ class GemmF16Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[256, 128, 64],
stages=3, warp_count=[4, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -283,13 +295,15 @@ class GemmF16Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -306,7 +320,7 @@ class GemmF16Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 64, 64],
stages=3, warp_count=[2, 1, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -324,13 +338,15 @@ class GemmF16Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -347,7 +363,7 @@ class GemmF16Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 256, 64],
stages=3, warp_count=[2, 4, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -365,13 +381,15 @@ class GemmF16Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -388,7 +406,7 @@ class GemmF16Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 256, 64],
stages=3, warp_count=[2, 4, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -406,13 +424,15 @@ class GemmF16Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
tools/library/scripts/pycutlass/test/gemm/gemm_f32_sm80.py
+15 -9
View File
@@ -19,7 +19,7 @@ class GemmF32nF32nF32nTensorOpF32Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 128, 32],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -37,13 +37,15 @@ class GemmF32nF32nF32nTensorOpF32Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -61,7 +63,7 @@ class GemmF32nF32nF32nTensorOpF32Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 128, 32],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -79,13 +81,15 @@ class GemmF32nF32nF32nTensorOpF32Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -102,7 +106,7 @@ class GemmF32nF32nF32nTensorOpF32Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[64, 64, 32],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -120,13 +124,15 @@ class GemmF32nF32nF32nTensorOpF32Sm80(unittest.TestCase):
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
tools/library/scripts/pycutlass/test/gemm/gemm_f64_sm80.py
+10 -6
View File
@@ -17,7 +17,7 @@ class GemmF64TensorOpSm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[32, 32, 16],
stages=4, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
# alignment 1 restricted for double
@@ -36,13 +36,15 @@ class GemmF64TensorOpSm80(unittest.TestCase):
element_epilogue = cutlass.float64
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -59,7 +61,7 @@ class GemmF64TensorOpSm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[64, 64, 16],
stages=4, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
# alignment 1 restricted for double
@@ -78,13 +80,15 @@ class GemmF64TensorOpSm80(unittest.TestCase):
element_epilogue = cutlass.float64
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
tools/library/scripts/pycutlass/test/gemm/gemm_grouped_sm80.py
+20 -16
View File
@@ -18,7 +18,7 @@ class GemmGroupedSm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 128, 32],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -37,14 +37,15 @@ class GemmGroupedSm80(unittest.TestCase):
)
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.BatchedIdentitySwizzle
for precompute_mode in [SchedulerMode.Device, SchedulerMode.Host]:
operation = GemmOperationGrouped(
tile_description.minimum_compute_capability,
80,
tile_description, A, B, C,
element_epilogue,
epilogue_functor, swizzling_functor,
precompute_mode=precompute_mode
)
@@ -64,7 +65,7 @@ class GemmGroupedSm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[64, 64, 16],
stages=4, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -83,14 +84,15 @@ class GemmGroupedSm80(unittest.TestCase):
)
element_epilogue = cutlass.float64
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.BatchedIdentitySwizzle
for precompute_mode in [SchedulerMode.Device, SchedulerMode.Host]:
operation = GemmOperationGrouped(
tile_description.minimum_compute_capability,
80,
tile_description, A, B, C,
element_epilogue,
epilogue_functor, swizzling_functor,
precompute_mode=precompute_mode
)
@@ -110,7 +112,7 @@ class GemmGroupedSm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 64, 8],
stages=4, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -129,14 +131,15 @@ class GemmGroupedSm80(unittest.TestCase):
)
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.BatchedIdentitySwizzle
for precompute_mode in [SchedulerMode.Device, SchedulerMode.Host]:
operation = GemmOperationGrouped(
tile_description.minimum_compute_capability,
80,
tile_description, A, B, C,
element_epilogue,
epilogue_functor, swizzling_functor,
precompute_mode=precompute_mode
)
@@ -156,7 +159,7 @@ class GemmGroupedSm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 128, 32],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -175,14 +178,15 @@ class GemmGroupedSm80(unittest.TestCase):
)
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.LinearCombination
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
swizzling_functor = cutlass.BatchedIdentitySwizzle
for precompute_mode in [SchedulerMode.Device, SchedulerMode.Host]:
operation = GemmOperationGrouped(
tile_description.minimum_compute_capability,
80,
tile_description, A, B, C,
element_epilogue,
epilogue_functor, swizzling_functor,
precompute_mode=precompute_mode
)
tools/library/scripts/pycutlass/test/gemm/gemm_s8_sm80.py
+28 -21
View File
@@ -1,5 +1,6 @@
import pycutlass
from pycutlass import *
from pycutlass.epilogue import LinearCombinationClamp
from pycutlass.test import *
import unittest
@@ -17,7 +18,7 @@ class GemmS8TensorOpF32Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[64, 64, 64],
stages=6, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -33,15 +34,15 @@ class GemmS8TensorOpF32Sm80(unittest.TestCase):
alignment=8
)
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.FastLinearCombinationClamp
epilogue_functor = FastLinearCombinationClamp(
C.element, C.alignment
)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -58,7 +59,7 @@ class GemmS8TensorOpF32Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 128, 128],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -74,15 +75,15 @@ class GemmS8TensorOpF32Sm80(unittest.TestCase):
alignment=16
)
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.FastLinearCombinationClamp
epilogue_functor = FastLinearCombinationClamp(
C.element, C.alignment
)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -99,7 +100,7 @@ class GemmS8TensorOpF32Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 128, 128],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -115,15 +116,15 @@ class GemmS8TensorOpF32Sm80(unittest.TestCase):
alignment=16
)
element_epilogue = cutlass.float32
epilogue_functor = EpilogueFunctor.FastLinearCombinationClamp
epilogue_functor = FastLinearCombinationClamp(
C.element, C.alignment
)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -140,7 +141,7 @@ class GemmS8TensorOpF32Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 128, 128],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -158,13 +159,16 @@ class GemmS8TensorOpF32Sm80(unittest.TestCase):
element_epilogue = cutlass.int32
epilogue_functor = EpilogueFunctor.LinearCombinationClamp
epilogue_functor = LinearCombinationClamp(
C.element, C.alignment, math_inst.element_accumulator,
element_epilogue
)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
@@ -181,7 +185,7 @@ class GemmS8TensorOpF32Sm80(unittest.TestCase):
tile_description = TileDescription(
threadblock_shape=[128, 128, 128],
stages=3, warp_count=[2, 2, 1],
math_instruction=math_inst, min_compute=80, max_compute=80
math_instruction=math_inst
)
A = TensorDescription(
@@ -199,13 +203,16 @@ class GemmS8TensorOpF32Sm80(unittest.TestCase):
element_epilogue = cutlass.int32
epilogue_functor = EpilogueFunctor.LinearCombinationClamp
epilogue_functor = LinearCombinationClamp(
C.element, C.alignment, math_inst.element_accumulator,
element_epilogue
)
swizzling_functor = cutlass.IdentitySwizzle1
operation = GemmOperationUniversal(
arch=80, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
tools/library/scripts/pycutlass/test/unit/cached_results_SM80_2080.txt
+13
View File
@@ -348,3 +348,16 @@ conv2d fprop_3x28x28x256_14x14_256x2x2_pad_h0w0_stride_h2w2_dil_h1w1_corr_alpha1
conv2d fprop_4x4x5x128_3x3_256x3x6_pad_h0w0_stride_h1w1_dil_h1w1_corr_alpha1_beta0 s8nhwc_s8nhwc_inhwc_i_i 164942943 4259285988 984016853 888753301
conv2d fprop_4x2x3x256_1x1_328x3x5_pad_h1w1_stride_h1w1_dil_h1w1_corr_alpha1_beta0 s8nhwc_s8nhwc_inhwc_i_i 2823094147 1681845497 4242738907 3244428635
conv2d fprop_1x17x11x288_17x11_160x3x3_pad_h1w1_stride_h1w1_dil_h1w1_corr_alpha2_beta2 s8nhwc_s8nhwc_inhwc_i_i 4060010502 2881035321 3927119619 3311661122
conv2d dgrad_1x11x7x64_6x4_8x1x1_pad_h0w0_stride_h2w2_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 4110991321 3464637181 1030377090 3211227145
conv2d dgrad_1x11x7x64_6x4_8x3x3_pad_h1w1_stride_h2w2_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 4110991321 1479940693 2379046159 2482639965
conv2d dgrad_1x13x11x64_8x7_8x1x1_pad_h1w1_stride_h2w2_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 832653836 1871463331 2718290800 1797658305
conv2d dgrad_1x17x19x64_9x10_16x2x2_pad_h1w1_stride_h2w2_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 3484040069 664160900 3954982568 985899371
conv2d dgrad_1x23x5x64_12x3_16x3x3_pad_h1w1_stride_h2w2_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 1513864544 1924855848 1728786974 3821277575
conv2d dgrad_1x55x51x256_28x26_512x1x1_pad_h0w0_stride_h2w2_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 868180534 1764715518 3998637379 2782670608
conv2d dgrad_1x27x23x256_9x7_512x3x3_pad_h0w0_stride_h3w3_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 3437976747 666906244 2107859856 831363691
conv2d dgrad_1x27x31x256_12x11_512x3x3_pad_h5w7_stride_h3w4_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 4195072693 1575210381 2486552517 3268706408
conv2d dgrad_1x27x35x256_15x9_512x7x5_pad_h11w7_stride_h3w5_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 3457330201 2316839359 1729888024 2308314800
conv2d dgrad_1x27x27x256_27x14_512x3x3_pad_h1w1_stride_h1w2_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 754609939 2469024119 464378888 544154978
conv2d dgrad_1x27x27x256_14x27_512x3x3_pad_h1w1_stride_h2w1_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 754609939 2469024119 464378888 3191247524
conv2d dgrad_3x28x28x256_14x14_256x2x2_pad_h0w0_stride_h2w2_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 1690216859 554790212 956712535 1281779197
conv2d dgrad_1x56x56x8_28x28_8x1x1_pad_h0w0_stride_h2w2_dil_h1w1_corr_alpha1_beta0 hnhwc_hnhwc_hnhwc_f_f 3184127693 835105643 4011933753 3207244654
tools/library/scripts/pycutlass/test/unit/test_sm80.py
+49 -38
View File
@@ -42,8 +42,7 @@ import unittest
#
def TestGemmOperator(gemm_kind, math_inst, layout, alignment, tiling, arch, mixed=False,
epilogue_functor = EpilogueFunctor.LinearCombination,
swizzling_functor=cutlass.IdentitySwizzle1, **kwargs):
epilogue_functor=None, swizzling_functor=cutlass.IdentitySwizzle1, **kwargs):
"""
Test GEMM Operation based on configuration
"""
@@ -68,7 +67,7 @@ def TestGemmOperator(gemm_kind, math_inst, layout, alignment, tiling, arch, mixe
tile_description = TileDescription(
tiling[0], tiling[1], tiling[2],
math_inst, arch, arch
math_inst
)
A = TensorDescription(
@@ -84,11 +83,15 @@ def TestGemmOperator(gemm_kind, math_inst, layout, alignment, tiling, arch, mixe
)
element_epilogue = data_type[3]
if epilogue_functor is None:
epilogue_functor = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, element_epilogue)
if gemm_kind == GemmKind.Universal:
operation = GemmOperationUniversal(
arch=arch, tile_description=tile_description,
A=A, B=B, C=C, element_epilogue=element_epilogue,
A=A, B=B, C=C,
epilogue_functor=epilogue_functor, swizzling_functor=swizzling_functor
)
if A.layout in [cutlass.ColumnMajorInterleaved32, cutlass.RowMajorInterleaved32]:
@@ -99,7 +102,7 @@ def TestGemmOperator(gemm_kind, math_inst, layout, alignment, tiling, arch, mixe
elif gemm_kind == GemmKind.Grouped:
operation = GemmOperationGrouped(
arch, tile_description, A, B, C,
element_epilogue, epilogue_functor, swizzling_functor,
epilogue_functor, swizzling_functor,
precompute_mode=kwargs["precompute_mode"]
)
testbed = TestbedGrouped(operation=operation)
@@ -110,7 +113,7 @@ def TestGemmOperator(gemm_kind, math_inst, layout, alignment, tiling, arch, mixe
def TestConv2dOperator(math_inst, alignment, tiling, arch,
stride_supports=[StrideSupport.Strided, StrideSupport.Strided, StrideSupport.Strided],
epilogue_functor=EpilogueFunctor.LinearCombination,
epilogue_functor=None,
swizzling_functor=cutlass.IdentitySwizzle1, interleaved=False, **kwargs):
"""
Test Conv2d Operation based on configurations
@@ -167,20 +170,24 @@ def TestConv2dOperator(math_inst, alignment, tiling, arch,
tile_description = TileDescription(
threadblock_shape=tiling[0], stages=tiling[1],
warp_count=tiling[2],
math_instruction=math_inst,
min_compute=arch, max_compute=arch
math_instruction=math_inst
)
if conv_kind == cutlass.conv.Operator.dgrad and stride_support == StrideSupport.Strided:
swizzling_functor = cutlass.StridedDgradIdentitySwizzle1
else:
swizzling_functor = default_swizzling_functor
if epilogue_functor is None:
epilogue_functor_ = LinearCombination(
C.element, C.alignment,
math_inst.element_accumulator, data_type[3])
operation = Conv2dOperation(
conv_kind=conv_kind, iterator_algorithm=cutlass.conv.IteratorAlgorithm.optimized,
arch=arch, tile_description=tile_description, A=A, B=B, C=C,
element_epilogue=data_type[3], stride_support=stride_support,
epilogue_functor=epilogue_functor,
stride_support=stride_support,
epilogue_functor=epilogue_functor_,
swizzling_functor=swizzling_functor
)
@@ -369,7 +376,11 @@ class Test_SM80(unittest.TestCase):
tiling = ([256, 64, 64], 4, [4, 1, 1])
data_type_mixed = [cutlass.int8, cutlass.int8, cutlass.int8, cutlass.float32]
self.assertTrue(TestGemmOperator(GemmKind.Universal, math_inst, layout, alignment_mixed, tiling, 80, False, data_type=data_type_mixed, epilogue_functor=EpilogueFunctor.FastLinearCombinationClamp))
epilogue_functor = FastLinearCombinationClamp(
data_type_mixed[2], alignment_mixed[2]
)
self.assertTrue(TestGemmOperator(GemmKind.Universal, math_inst, layout, alignment_mixed, tiling, 80, False, data_type=data_type_mixed, epilogue_functor=epilogue_functor))
stride_supports = [StrideSupport.Strided, StrideSupport.Strided, StrideSupport.Strided]
layout = [cutlass.TensorNC32HW32, cutlass.TensorC32RSK32, cutlass.TensorNC32HW32]
results = TestConv2dOperator(math_inst, alignment_mixed, tiling, 80, stride_supports=stride_supports, data_type=data_type_mixed, layout=layout, interleaved=True)
@@ -378,59 +389,59 @@ class Test_SM80(unittest.TestCase):
def SM80_SparseTensorOp_16832(self):
pass
def test_SM80_PlanarComplexTensorOp_16816(self):
def SM80_PlanarComplexTensorOp_16816(self):
pass
def test_SM80_SparseTensorOp_16816_fast_math(self):
def SM80_SparseTensorOp_16816_fast_math(self):
pass
def test_SM80_TensorOp_1688_complex(self):
def SM80_TensorOp_1688_complex(self):
pass
def test_SM80_TensorOp_1688_fast_fp32_math_complex(self):
def SM80_TensorOp_1688_fast_fp32_math_complex(self):
pass
def test_SM80_TensorOp_1688_rank_k(self):
def SM80_TensorOp_1688_rank_k(self):
pass
def test_SM80_TensorOp_1688_rank_k_complex(self):
def SM80_TensorOp_1688_rank_k_complex(self):
pass
def test_SM80_TensorOp_1688_trmm(self):
def SM80_TensorOp_1688_trmm(self):
pass
def test_SM80_TensorOp_1688_trmm_complex(self):
def SM80_TensorOp_1688_trmm_complex(self):
pass
def test_SM80_TensorOp_1688_symm(self):
def SM80_TensorOp_1688_symm(self):
pass
def test_SM80_TensorOp_1688_symm_complex(self):
def SM80_TensorOp_1688_symm_complex(self):
pass
def test_SM80_TensorOp_884_complex(self):
def SM80_TensorOp_884_complex(self):
pass
def test_SM80_TensorOp_884_complex_gaussian(self):
def SM80_TensorOp_884_complex_gaussian(self):
pass
def test_SM80_TensorOp_884_rank_k(self):
def SM80_TensorOp_884_rank_k(self):
pass
def test_SM80_TensorOp_884_rank_k_complex(self):
def SM80_TensorOp_884_rank_k_complex(self):
pass
def test_SM80_TensorOp_884_rank_k_complex_gaussian(self):
def SM80_TensorOp_884_rank_k_complex_gaussian(self):
pass
def test_SM80_TensorOp_884_trmm(self):
def SM80_TensorOp_884_trmm(self):
pass
def test_SM80_TensorOp_884_trmm_complex(self):
def SM80_TensorOp_884_trmm_complex(self):
pass
def test_SM80_TensorOp_884_trmm_complex_gaussian(self):
def SM80_TensorOp_884_trmm_complex_gaussian(self):
pass
def test_SM80_TensorOp_884_symm(self):
def SM80_TensorOp_884_symm(self):
pass
def test_SM80_TensorOp_884_symm_complex(self):
def SM80_TensorOp_884_symm_complex(self):
pass
def test_SM80_TensorOp_884_symm_complex_gaussian(self):
def SM80_TensorOp_884_symm_complex_gaussian(self):
pass
def test_SM80_SparseTensorOp_16864_TN(self):
def SM80_SparseTensorOp_16864_TN(self):
pass
def test_SM80_TensorOp_16864_TN(self):
def SM80_TensorOp_16864_TN(self):
pass
def test_SM80_SparseTensorOp_168128_TN(self):
def SM80_SparseTensorOp_168128_TN(self):
pass
def test_SM80_TensorOp_16864_Interleaved(self):
def SM80_TensorOp_16864_Interleaved(self):
pass
def test_SM80_TensorOp_168256(self):
def SM80_TensorOp_168256(self):
pass
def test_SM80_Simt_complex(self):
def SM80_Simt_complex(self):
pass