]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 1a87dcd)
ggml-cpu: fix SVE leftover path in ggml_vec_dot_f32 (#24699)
authorTarek Dakhran <redacted>
Fri, 26 Jun 2026 07:41:56 +0000 (09:41 +0200)
committerGitHub <redacted>
Fri, 26 Jun 2026 07:41:56 +0000 (10:41 +0300)
commitc16c35b8142ac34b4559a5e8584e8daaf3958416
tree253cf542a0577beb8ad033a193c264ba2ef05d8ctree
parent1a87dcdc452de056a66796b679e4ba7071488cf7commit | diff
ggml-cpu: fix SVE leftover path in ggml_vec_dot_f32 (#24699)

* ggml-cpu: fix SVE leftover path in ggml_vec_dot_f32

2D convolutions with kernel size 9 produced different results on SVE
enabled ARM devices. After debugging it turned out that ggml_vec_dot_f32
was using data from inactive lanes.

Use svmla_f32_m(pg, sum1, ax1, ay1) so inactive lanes retain sum1.

* cont : clean-up

---------

Co-authored-by: Georgi Gerganov <redacted>
ggml/src/ggml-cpu/vec.cpp diff | blob | history
tests/test-quantize-fns.cpp diff | blob | history
Packaging of ggml-org/llama.cpp