Piotr Wilkin (ilintar)andJohannes Gäßler 61b318f44c CUDA/HIP: Flash Attention tuning (gfx1201) (llama/28102) ...

* HIP: enable mma FA for head size 256 on RDNA4, tune configs

Assisted-by: Claude
Assisted-by: Codex

* HIP: prefer whole-tile FA grids over stream-k on AMD WMMA

Assisted-by: Claude
Assisted-by: Codex

* revise stream_k logic

* revise kernel selection logic

---------

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

2026-09-14 20:45:06 +03:00

..

cmake

CUDA: replace GGML_FA_ALL_QUANTS with GGML_FA_QUANTS, more control over what is compiled (llama/28079)

2026-09-14 20:45:06 +03:00

include

ggml : update ggml_prec specification (llama/26675)

2026-09-14 20:45:06 +03:00

src

CUDA/HIP: Flash Attention tuning (gfx1201) (llama/28102)

2026-09-14 20:45:06 +03:00

.gitignore

…

CMakeLists.txt

CUDA: replace GGML_FA_ALL_QUANTS with GGML_FA_QUANTS, more control over what is compiled (llama/28079)

2026-09-14 20:45:06 +03:00