]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
opencl: Adreno optimization for MoE - MxFP4 (llama/22301)
authorShawn Gu <redacted>
Sat, 2 May 2026 06:02:24 +0000 (23:02 -0700)
committerGeorgi Gerganov <redacted>
Thu, 14 May 2026 18:26:48 +0000 (21:26 +0300)
commitff5704a416813610e30e54d864c5af1be41288c6
treee0b788642e2c225ef9f5252c8ce9ef5667b76bdb
parent3e9b7d0fef3528ee2208da3cdb873a2c53d2ae2f
opencl: Adreno optimization for MoE - MxFP4 (llama/22301)

* MoE Mxfp4 CLC kernel added, router reorder on GPU

* Pass test-backend-ops for MoE mxfp4 Adreno CLC

* remove putenv in llama-model.cpp

* fix indent style and whitespace

* opencl: remove unnecessary headers

* opencl: do not save cl_program objects

* opencl: remove unnecessary assert

* fix precision issue

---------

Co-authored-by: Li He <redacted>
ggml/src/ggml-opencl/CMakeLists.txt
ggml/src/ggml-opencl/ggml-opencl.cpp
ggml/src/ggml-opencl/kernels/cvt.cl
ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32_ns.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32_ns.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/moe_reorder_b.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/moe_sort_by_expert.cl [new file with mode: 0644]