leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 26 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
)