leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 29 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/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
)