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

vulkan: use 4 rows for scalar FA large tile size

This commit is contained in:
Jeff Bolz 2025-12-14 11:15:49 -06:00
parent 5266379bca
commit cdb09db91e
1 changed files with 2 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -2535,12 +2535,11 @@ static constexpr uint32_t flash_attention_num_small_rows = 32;
static constexpr uint32_t scalar_flash_attention_num_small_rows = 1; static constexpr uint32_t scalar_flash_attention_num_small_rows = 1;
static uint32_t get_fa_scalar_num_large_rows(uint32_t hsk, uint32_t hsv) { static uint32_t get_fa_scalar_num_large_rows(uint32_t hsk, uint32_t hsv) {
GGML_UNUSED(hsk);
if (hsv >= 192) { if (hsv >= 192) {
return 2; return 2;
} else if ((hsv | hsk) & 8) {
return 4;
} else { } else {
return 8; return 4;
} }
} }