CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: use scalar FA rather than coopmat2 when N==1 (#13554)

Browse Source
This commit is contained in:
Jeff Bolz
2025-05-17 15:35:47 +09:00
committed by GitHub
parent 3e0be1cace
commit 4f41ee11d6
1 changed files with 7 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@@ -5872,10 +5872,17 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx
vk_pipeline *pipelines; vk_pipeline *pipelines;
bool small_rows = N <= get_fa_num_small_rows(path); bool small_rows = N <= get_fa_num_small_rows(path);
// coopmat1 does not actually support "small rows" (it needs 16 rows).
// So use scalar instead.
if (small_rows && path == FA_COOPMAT1) { if (small_rows && path == FA_COOPMAT1) {
path = FA_SCALAR; path = FA_SCALAR;
} }
// scalar is faster than coopmat2 when N==1
if (N == 1 && path == FA_COOPMAT2) {
path = FA_SCALAR;
}
bool f32acc = path == FA_SCALAR || dst->op_params[3] == GGML_PREC_F32; bool f32acc = path == FA_SCALAR || dst->op_params[3] == GGML_PREC_F32;
switch (path) { switch (path) {