]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
opencl: add q8_0 mm support (llama/16469)
authorlhez <redacted>
Wed, 15 Oct 2025 17:51:04 +0000 (10:51 -0700)
committerGeorgi Gerganov <redacted>
Tue, 21 Oct 2025 15:14:33 +0000 (18:14 +0300)
commit4560236227d6b561390940c113a6523b54d73b26
treefc4139ff102431803103c6f5e4ecd5b282ea4ce4
parent70e35abfce31240e36921dd2b7c4a34867651fe0
opencl: add q8_0 mm support (llama/16469)

* opencl: add mm_q8_0_f32

* opencl: fix data loading for incomplete tile

* opencl: use q8_0 mm for larger matrix

* opencl: add some tests to cover the path
src/ggml-opencl/CMakeLists.txt
src/ggml-opencl/ggml-opencl.cpp
src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl
src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl
src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl [new file with mode: 0644]
tests/test-backend-ops.cpp