]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: ac2557c)
opencl: loads quants as uint for q4_K and q5_K flat mv (optimization for Adreno A7x...
authorHongqiang Wang <redacted>
Thu, 16 Jul 2026 20:18:21 +0000 (13:18 -0700)
committerGitHub <redacted>
Thu, 16 Jul 2026 20:18:21 +0000 (13:18 -0700)
commite8f19cc0ad70a243c8012bf17b4be601abfc8ea2
treecf9d76444980134269463c777d804e4d3e27a4a2tree
parentac2557cb24def295888ef47f1a35b401d978c510commit | diff
opencl: loads quants as uint for q4_K and q5_K flat mv (optimization for Adreno A7x GPUs) (#25780)

* opencl: load quant as uint in mv_q4_k_f32_flat

helps older compilers (e.g., E031.41, boosts 2x),
no impact on newer compilers (e.g., E031.45 or newer)

* opencl: load quant as uint in mv_q5_K_f32_flat

helps older compilers (e.g., E031.41)

* opencl: format

---------

Co-authored-by: Li He <redacted>
ggml/src/ggml-opencl/kernels/mul_mv_q4_k_f32_flat.cl diff | blob | history
ggml/src/ggml-opencl/kernels/mul_mv_q5_k_f32_flat.cl diff | blob | history
Packaging of ggml-org/llama.cpp