happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

use smaller scalar rows size for smaller rows count

This commit is contained in:
Ruben Ortlam 2026-02-10 18:23:43 +01:00
parent de6db3fed6
commit d6a004547f
1 changed files with 1 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -2769,9 +2769,7 @@ static uint32_t get_fa_scalar_num_rows(uint32_t hsk, uint32_t hsv, FaRows rows,
return 8;
}
if (hsv >= 192) {
return 8;
} else if ((hsv | hsk) & 8 || small_cache) {
if (hsv >= 192 || (hsv | hsk) & 8 || small_cache || rows == FA_ROWS_2 || rows == FA_ROWS_4 || rows == FA_ROWS_8) {
return 8;
}