]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
opencl: add int8 dp4 dense and MoE prefill optimization for Adreno GPUs (#25537)
authorHongqiang Wang <redacted>
Sat, 11 Jul 2026 06:05:58 +0000 (23:05 -0700)
committerGitHub <redacted>
Sat, 11 Jul 2026 06:05:58 +0000 (23:05 -0700)
commit1d1d9a9ed7a4f09c4225ea4cc8fd3bd1cf2c940f
treeb8c731d4146dcbfe1a1b711b2dda7379594664db
parent4f37f519722aa3242eecb7649466b4a4a2d6d6da
opencl: add int8 dp4 dense and MoE prefill optimization for Adreno GPUs (#25537)

* opencl: add int8 dp4 dense and moe GEMM

* opencl: refactor

---------

Co-authored-by: Li He <redacted>
21 files changed:
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_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_moe_q4_0_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_moe_q4_k_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_moe_q6_k_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_moe_q8_0_f32_ns.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_moe_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_noshuffle_iq4_nl_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_0_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_k_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_0_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_k_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_noshuffle_q6_k_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemm_noshuffle_q8_0_q8_1_dp4a.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32_ns.cl
ggml/src/ggml-opencl/kernels/gemv_moe_q4_k_f32_ns.cl
ggml/src/ggml-opencl/kernels/moe_combine.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/moe_reorder_quant_a_q8_1.cl [new file with mode: 0644]
ggml/src/ggml-opencl/kernels/quant_a_q8_1.cl [new file with mode: 0644]