ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-12 02:05:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix the build flag for moe blockscale gfx950.

Browse Source
This commit is contained in:
mtgu0705
2025-05-03 08:48:58 -05:00
parent 6a0d86f683
commit f27fbdee49
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
example/65_gemm_multiply_multiply/CMakeLists.txt
View File

@@ -7,7 +7,7 @@ add_example_executable(example_moe_gemm1_xdl_fp8 moe_gemm1_xdl_fp8.cpp)
add_example_executable(example_moe_gemm2_xdl_fp8 moe_gemm2_xdl_fp8.cpp)
add_example_executable(example_moe_gemm2_xdl_fp8_blockscale moe_gemm2_xdl_fp8_blockscale.cpp)
list(APPEND gpu_list gfx942)
list(APPEND gpu_list gfx942 gfx950)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
@@ -18,5 +18,5 @@ foreach(gpu IN LISTS GPU_TARGETS)
endforeach()
set(EXAMPLE_COMPILE_OPTIONS)
list(APPEND EXAMPLE_COMPILE_OPTIONS -v --save-temps -Wno-gnu-line-marker)
list(APPEND EXAMPLE_COMPILE_OPTIONS "SHELL: -mllvm -greedy-reverse-local-assignment=1 -mllvm --schedmodel=0 -mllvm -misched=gcn-iterative-max-occupancy-experimental")
list(APPEND EXAMPLE_COMPILE_OPTIONS "SHELL: -mllvm -greedy-reverse-local-assignment=1 -mllvm --schedmodel=0 -mllvm --amdgpu-sched-strategy=gcn-iterative-max-occupancy-experimental")
target_compile_options(example_moe_gemm2_xdl_fp8_blockscale PRIVATE ${EXAMPLE_COMPILE_OPTIONS})

4
example/65_gemm_multiply_multiply/moe_gemm2_xdl_fp8_blockscale.cpp
View File

@@ -321,9 +321,9 @@ int main(int argc, char* argv[])
{
case 0: break;
case 1:
a0_t_k_k.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-2, 2});
a0_t_k_k.GenerateTensorValue(GeneratorTensor_3<A0DataType>{-0.5, 0.5});
a1_t_k_k.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0, 1.0});
b0_e_n_k.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
b0_e_n_k.GenerateTensorValue(GeneratorTensor_3<B0DataType>{-0.5, 0.5});
b1_e_n_k.GenerateTensorValue(GeneratorTensor_3<B1DataType>{0, 1.0});
d2_e_n.GenerateTensorValue(GeneratorTensor_3<D2DataType>{0, 1.0});
break;