ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-29 03:07:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

Modify the a_thread offset since the A data load is different from B.

Browse Source
This commit is contained in:
mtgu0705
2024-12-25 23:26:17 +08:00
parent 1fcd332967
commit f728087c61
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_multi_d_ab_scale.hpp
View File

@@ -1368,8 +1368,10 @@ struct GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3
make_tuple(Number<ScaleSliceSizeN>{}, Number<ScaleSliceSizeK>{}));
constexpr index_t MWaves = MPerBlock / (MXdlPerWave * MPerXdl);
auto a_thread_offset =
get_thread_local_1d_id() % MPerXdl + (get_thread_local_1d_id() / 64) % MWaves * MPerXdl;
// auto a_thread_offset =
// get_thread_local_1d_id() % MPerXdl + (get_thread_local_1d_id() / 64) % MWaves * MPerXdl;
auto a_thread_offset = get_thread_local_1d_id() % MPerXdl + (get_thread_local_1d_id() / 128) * MPerXdl;
auto a_scale_thread_copy =
ThreadwiseTensorSliceTransfer_v2<AScaleType,