msccl: build same number of kernels as in ROCm 5.7 (#1005)

Removed fullOps kernels from build
This commit is contained in:
Wenkai Du authored and GitHub committed 2023-12-07 13:36:04 -06:00
1 parent 9c3fea1751
commit 12c08fc52a
5 files changed
+18 -29

No files matched your search

+3 -7
View File
@@ -39,14 +39,10 @@ __global__ void MSCCL_KERNEL_ENTRY_NAME(devredop, type, proto, fullOps)(struct n
#define MSCCL_DECL_KERNEL_ENTRY_FUNC() \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(Sum, false) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(Prod, false) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(Max, false) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(Min, false) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(Sum, true) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(Prod, true) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(Max, true) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(Min, true) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(PreMulSum, true) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP_NOFLOAT(SumPostDiv, true)
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(Max, false) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP(PreMulSum, false) \
MSCCL_DECL_KERNEL_ENTRY_FUNC_DEVREDOP_NOFLOAT(SumPostDiv, false)
MSCCL_DECL_KERNEL_ENTRY_FUNC()