]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: d8eaa26)
ggml : Leverage the existing GGML_F32_VEC helpers to vectorize ggml_vec_set_f32 for...
authorsirus20x6 <redacted>
Wed, 22 Oct 2025 10:14:14 +0000 (05:14 -0500)
committerGitHub <redacted>
Wed, 22 Oct 2025 10:14:14 +0000 (12:14 +0200)
commit19a5a3edfd306516cc419679d69d6435943b6816
tree628625ae8df3ab265d6d01500ad4e289ded285f7tree
parentd8eaa26e4d9228df3aa46a930db60c8eaab67c1bcommit | diff
ggml : Leverage the existing GGML_F32_VEC helpers to vectorize ggml_vec_set_f32 for faster fills (#16522)

* Leverage the existing GGML_F32_VEC helpers to broadcast the fill value across SIMD registers and store in vector-sized chunks, while retaining the scalar tail for leftover elements and non-SIMD builds.

* Vectorize additional f32 helper loops

* Normalize f32 helper tails for ggml vec ops

---------

Co-authored-by: Aaron <redacted>
ggml/src/ggml-cpu/vec.h diff | blob | history
Packaging of ggml-org/llama.cpp