ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-17 17:19:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

reproduce

Browse Source
This commit is contained in:
valarLip
2025-08-06 10:09:33 +00:00
parent 07469142cb
commit 5a0219b7c7
2 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/65_gemm_multiply_multiply/CMakeLists.txt
View File

@@ -40,6 +40,8 @@ foreach(gpu IN LISTS GPU_TARGETS)
endforeach()
set(GEMM_OPTIONS "SHELL: -mllvm -greedy-reverse-local-assignment=1")
list(APPEND GEMM_OPTIONS -Wno-error --save-temps)
set(BLOCKSCALE_GEMM_OPTIONS )
check_cxx_compiler_flag("-mllvm --misched-bottomup=1" HAS_MISCHED_BOTTOMUP)
check_cxx_compiler_flag("-mllvm --misched-prera-direction=bottomup" HAS_MISCHED_PRERA_DIRECTION)

2
example/65_gemm_multiply_multiply/moe_gemm1_xdl_fp8.cpp
View File

@@ -157,7 +157,7 @@ using AElementOp = PassThrough;
using BElementOp = PassThrough;
static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::Default;
static constexpr ck::index_t MPerBlock = 128;
static constexpr ck::index_t MPerBlock = 256;
static constexpr ck::index_t NPerBlock = 128;
static constexpr ck::index_t MNPerXDL = 16;
static constexpr ck::index_t MXDLPerWave = MPerBlock / (MNPerXDL * 1);