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_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
)