]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: fd89556)
SYCL: reduce allocation overhead during flash attention (#22732)
authorAlexey Kopytko <redacted>
Sat, 9 May 2026 06:30:39 +0000 (15:30 +0900)
committerGitHub <redacted>
Sat, 9 May 2026 06:30:39 +0000 (09:30 +0300)
commite20b83930cc0a1cefaaa81774ca1a0fcfc76964c
tree453122bc6bde8a4540c29eccc6c779c9a975313btree
parentfd89556567057bf64a6f6d6e50abec488929d7e0commit | diff
SYCL: reduce allocation overhead during flash attention (#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
docs/backend/SYCL.md diff | blob | history
ggml/src/ggml-sycl/common.hpp diff | blob | history
ggml/src/ggml-sycl/fattn-buffers.cpp [new file with mode: 0644] blob
ggml/src/ggml-sycl/fattn-buffers.hpp [new file with mode: 0644] blob
ggml/src/ggml-sycl/fattn-common.hpp diff | blob | history
ggml/src/ggml-sycl/ggml-sycl.cpp diff | blob | history
Packaging of ggml-org/llama.cpp