]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
opencl: add optimized q4_1 mm kernel for adreno (llama/19840)
authorshaofeiqi <redacted>
Tue, 3 Mar 2026 03:49:41 +0000 (19:49 -0800)
committerGeorgi Gerganov <redacted>
Sun, 15 Mar 2026 19:50:13 +0000 (21:50 +0200)
commite322754191d7a18c3f6c5706f4aa6a2deb19765b
treec0daed6a5d0be743aa6d072112e4bbc54ce7a41e
parent18efe2c5527dca44bfcaa93e4dbf0ed7df07563a
opencl: add optimized q4_1 mm kernel for adreno (llama/19840)

* Add Q4_1 OpenCL Kernels

* opencl: refactor transpose

* opencl: format

* opencl: refactor q4_1 unpack

* opencl: move `ggml_cl_mul_mat_q4_1_f32_adreno`

* opencl: refactor `ggml_cl_mul_mat_q4_1_f32_adreno` and kernels

* opencl: rename kernel files and kernes

* opencl: fix build for non adreno

* opencl: move code around and format

---------

Co-authored-by: Li He <redacted>
src/ggml-opencl/CMakeLists.txt
src/ggml-opencl/ggml-opencl.cpp
src/ggml-opencl/kernels/cvt.cl
src/ggml-opencl/kernels/gemm_noshuffle_q4_1_f32.cl [new file with mode: 0644]
src/ggml-opencl/kernels/gemv_noshuffle_general_q8_0_f32.cl
src/ggml-opencl/kernels/gemv_noshuffle_q4_1_f32.cl [new file with mode: 0644]
src/ggml-opencl/kernels/transpose.cl