]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
vulkan: preprocess mul_mat_id experts and discard workgroups more quickly (#18352)
authorJeff Bolz <redacted>
Fri, 26 Dec 2025 22:12:58 +0000 (16:12 -0600)
committerGitHub <redacted>
Fri, 26 Dec 2025 22:12:58 +0000 (16:12 -0600)
commitc9ced4910ba0c8d95565950ac11bcc54fee309cd
treec9a3958eb8a293413152d1f59f09f3fe43aadc0f
parent7ac8902133da6eb390c4d8368a7d252279123942
vulkan: preprocess mul_mat_id experts and discard workgroups more quickly (#18352)

Run a preprocess to count how many times each expert is used, and use this to
quickly discard workgroups that aren't needed.
ggml/src/ggml-vulkan/ggml-vulkan.cpp
ggml/src/ggml-vulkan/vulkan-shaders/count_experts.comp [new file with mode: 0644]
ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp
ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp
ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_id_funcs.glsl
ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp