DeepSeek/DeepGEMM
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepGEMM synced 2025-05-06 15:44:23 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Minor fixes

Browse Source
This commit is contained in:
Chenggang Zhao 2025-04-23 11:38:45 +08:00
parent 07ef809d82
commit 55d1d01c43
2 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
deep_gemm/include/deep_gemm/fp8_gemm.cuh
View File

@ -226,8 +226,8 @@ fp8_gemm_kernel(__nv_bfloat16* gmem_d, float* scales_b, int* grouped_layout,
// Wait unaligned cases
#pragma unroll
for (uint32_t s = kNumInnerStages; s < kNumStages; ++ s) {
empty_barriers[s]->wait((scheduler.current_iter * kNumIterations + k_iter + 1) & 1);
full_barriers[s]->arrive();
empty_barriers[s]->wait((scheduler.current_iter * kNumIterations + k_iter + 1) & 1);
}
}, 0);
}

2
deep_gemm/include/deep_gemm/scheduler.cuh
View File

@ -118,7 +118,7 @@ struct Scheduler {
if (curr_group_idx == kNumGroups)
return false;
// Within current group
// Within the current group
num_m_blocks = ceil_div(static_cast<uint32_t>(__ldg(grouped_layout + curr_group_idx)), BLOCK_M);
auto current_m_block_cumsum = curr_cumsum + num_m_blocks;
if (next_block_idx < current_m_block_cumsum * kNumNBlocks)