DeepSeek/DeepGEMM
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepGEMM synced 2025-04-03 13:00:45 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Add some notes

Browse Source
This commit is contained in:
Chenggang Zhao 2025-03-25 17:41:49 +08:00
parent 25db8de345
commit 09d097f84d
1 changed files with 1 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
deep_gemm/include/deep_gemm/fp8_gemm.cuh
View File

@ -314,6 +314,7 @@ fp8_gemm_kernel(__nv_bfloat16* gmem_d, float* scales_b, int* grouped_layout,
#pragma unroll
for (int i = 0; i < WGMMA::kNumAccum / 4; ++ i) {
// NOTES: for unrolled `num_former_iters` cases, we expect the compiler to automatically make it a constant
bool predicate = kMustUseUniformedScaleB or i < num_former_iters;
final_accum[i * 4 + 0] += (predicate ? scale_0_0 : scale_0_1) * accum[i * 4 + 0];
final_accum[i * 4 + 1] += (predicate ? scale_0_0 : scale_0_1) * accum[i * 4 + 1];