DeepSeek/DeepGEMM
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepGEMM synced 2025-05-08 13:44:37 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Add some notes for promotion

Browse Source
This commit is contained in:
Chenggang Zhao 2025-03-04 11:42:20 +08:00
parent ded740f736
commit 9b0dad8640
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

@ -283,6 +283,7 @@ fp8_gemm_kernel(__nv_bfloat16* gmem_d, float* scales_b, int* grouped_layout,
empty_barrier_arrive(s); empty_barrier_arrive(s);
// Promote with scales // Promote with scales
// NOTES: making it as predicates is very important for performance, comparing to two loops
float scale_0_0 = scale_a_0 * scale_b_0, scale_1_0 = scale_a_1 * scale_b_0; float scale_0_0 = scale_a_0 * scale_b_0, scale_1_0 = scale_a_1 * scale_b_0;
float scale_0_1, scale_1_1; float scale_0_1, scale_1_1;
if constexpr (not kMustUseUniformedScaleB) if constexpr (not kMustUseUniformedScaleB)