This commit is contained in:
coderfeli
2025-04-04 13:40:10 +00:00
parent 491178276f
commit 867a4e527c
2 changed files with 2 additions and 1 deletions

View File

@@ -314,6 +314,7 @@ struct BlockwiseGemmXdlops_pipeline_bpreshuffle_gufusion_bdequant_v1<
// Initialize C
c_thread_buf.Clear();
c_thread_buf_up.Clear();
__builtin_amdgcn_sched_barrier(0);

View File

@@ -1322,7 +1322,7 @@ struct GridwiseMoeGemm
KPerBlock);
if constexpr(IsInputGemm)
{
const BDataType* p_b_grid_up = p_b_grid + expert_stride / 2;
const BDataType* p_b_grid_up = p_b_grid + expert_stride / 2 / BPackedSize;
const auto b_grid_buf_up = make_dynamic_buffer<AddressSpaceEnum::Global>(
p_b_grid_up + expert_id * expert_stride / BPackedSize,
b_grid_desc_bpreshuffled.GetElementSpaceSize());