]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
SYCL: reduce allocation overhead during flash attention (llama/22732)
authorAlexey Kopytko <redacted>
Sat, 9 May 2026 06:30:39 +0000 (15:30 +0900)
committerGeorgi Gerganov <redacted>
Thu, 14 May 2026 18:26:48 +0000 (21:26 +0300)
commit8c7efe885cb38b9617ca952ebf7bc19d79bd6ffa
tree39bf8ef9b71c94f1746380e9931cfc71501ad96d
parent25f543175d0652204eacd643864a09a8b5fd39fe
SYCL: reduce allocation overhead during flash attention (llama/22732)

* SYCL: reduce allocation overhead during flash attention

* tidy up whitespace

* add a note about the flag

* move ggml_sycl_fattn_* into fattn-buffers.hpp

* refactor implementation into fattn-buffers.cpp

* move new_fattn_kv_buffers back into ggml-sycl.cpp
ggml/src/ggml-sycl/common.hpp
ggml/src/ggml-sycl/fattn-buffers.cpp [new file with mode: 0644]
ggml/src/ggml-sycl/fattn-buffers.hpp [new file with mode: 0644]
ggml/src/ggml-sycl/fattn-common.hpp
ggml/src/ggml-sycl/ggml-sycl.cpp