fix instruction seq

This commit is contained in:
coderfeli
2025-02-19 11:04:21 +00:00
parent a41f76cdea
commit 76644d70dd

View File

@@ -182,9 +182,9 @@ struct BlockwiseGemmXdlops_pipeline_bpreshuffle_v1<BlockGemmPipelineScheduler::I
__device__ static constexpr auto HotLoopScheduler()
{
constexpr auto num_ds_read_inst_a = HotLoopInstList::A_LDS_Read_Inst_Num;
constexpr auto num_ds_read_inst_a = HotLoopInstList::A_LDS_Read_Inst_Num * 2;
constexpr auto num_buffer_load_inst_a = HotLoopInstList::A_Buffer_Load_Inst_Num;
constexpr auto num_buffer_load_inst_b = HotLoopInstList::B_Buffer_Load_Inst_Num;
constexpr auto num_buffer_load_inst_b = HotLoopInstList::B_Buffer_Load_Inst_Num * 2;
// B global
static_for<0, num_buffer_load_inst_b, 1>{}([&](auto i) {