ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-11 09:40:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

find the root cause error in did not enable the transpose in gfx950 correctly

Browse Source
This commit is contained in:
ThomasNing
2025-09-23 03:36:04 -05:00
parent cbffc459fd
commit e2a0df59ea
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
include/ck_tile/ops/gemm/pipeline/gemm_pipeline_ag_bg_cr_comp_v3.hpp
View File

@@ -542,7 +542,7 @@ struct GemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Problem>
{
block_sync_lds();
if constexpr(is_a_col_major)
if constexpr(is_a_col_major && !is_a_load_tr_v())
{
auto a_shuffle_tmp = make_static_distributed_tensor<ADataType>(
Policy::template MakeShuffledARegTileDistribution<Problem>());
@@ -553,7 +553,7 @@ struct GemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Problem>
{
Base::LocalPrefill(a_copy_lds_window, elementwise_As_res);
}
if constexpr(is_b_row_major)
if constexpr(is_b_row_major && !is_b_load_tr_v())
{
auto b_shuffle_tmp = make_static_distributed_tensor<BDataType>(
Policy::template MakeShuffledBRegTileDistribution<Problem>());
@@ -577,7 +577,7 @@ struct GemmPipelineAgBgCrCompV3 : public BaseGemmPipelineAgBgCrCompV3<Problem>
block_sync_lds();
block_gemm.LocalPrefetch(a_lds_gemm_window, b_lds_gemm_window);
block_gemm.LocalPrefetch(a_lds_gemm_window, b_lds_gemm_window, is_a_load_tr_v, is_b_load_tr_v);
HotLoopScheduler();
__builtin_amdgcn_sched_barrier(0);