Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 06:10:29 +01:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: remove unnecessary warp reduce in FA (ggml/1032)

Browse Source 
* kqmax_new_j in every thread within warp is same after operate at line 199,this reduce can be omit

* same problem in vec32

---------

Co-authored-by: ZhaoXiaoYu <zhao.xiaoyu@zte.com.cn>
This commit is contained in:
mahorozte 2024-12-03 21:11:43 +08:00 committed by Georgi Gerganov
parent efb6ae9630
commit e9e661bd59
2 changed files with 0 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
ggml/src/ggml-cuda/fattn-vec-f16.cuh
View File

@ -220,7 +220,6 @@ static __global__ void flash_attn_vec_ext_f16(
for (int j = 0; j < ncols; ++j) {
half kqmax_new_j = ncols == 1 ? kqmax_new : kqmax_new_arr[j];
kqmax_new_j = warp_reduce_max(kqmax_new_j);
if (threadIdx.x == 0) {
kqmax_shared[j][threadIdx.y] = kqmax_new_j;
}

1
ggml/src/ggml-cuda/fattn-vec-f32.cuh
View File

@ -206,7 +206,6 @@ static __global__ void flash_attn_vec_ext_f32(
for (int j = 0; j < ncols; ++j) {
float kqmax_new_j = kqmax_new_arr[j];
kqmax_new_j = warp_reduce_max(kqmax_new_j);
if (threadIdx.x == 0) {
kqmax_shared[j][threadIdx.y] = kqmax_new_j;
}