Skip to content

CUDA: fix kernel selection logic for tile FA #19686

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
JohannesGaessler merged 2 commits into from
Feb 19, 2026
Merged

CUDA: fix kernel selection logic for tile FA #19686

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
566971b
CUDA: fix kernel selection logic for tile FA
JohannesGaessler Feb 17, 2026
3fca548
add comment [no ci]
JohannesGaessler Feb 19, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion ggml/src/ggml-cuda/fattn-tile.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1186,8 +1186,10 @@ static void launch_fattn_tile_switch_ncols2(ggml_backend_cuda_context & ctx, ggm
GGML_ASSERT(Q->ne[2] % K->ne[2] == 0);
const int gqa_ratio = Q->ne[2] / K->ne[2];

// On NVIDIA (Pascal and older) the GQA optimizations seem to be detrimental in some cases.
// However, for DKQ == 576, DV == 512 only the kernel variant with GQA optimizations is implemented.
const bool nvidia = GGML_CUDA_CC_IS_NVIDIA(ggml_cuda_info().devices[ggml_cuda_get_device()].cc);
const int gqa_limit = nvidia && gqa_ratio <= 4 ? 16 : INT_MAX;
const int gqa_limit = nvidia && gqa_ratio <= 4 && DV <= 256 ? 16 : INT_MAX;
const bool use_gqa_opt = mask && max_bias == 0.0f && Q->ne[1] <= gqa_limit && K->ne[1] % FATTN_KQ_STRIDE == 0;

if constexpr (DV == 512) {
Expand Down