]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
vulkan: Use larger loads in scalar/coopmat1 matmul (llama/15729)
authorJeff Bolz <redacted>
Sun, 7 Sep 2025 16:53:07 +0000 (11:53 -0500)
committerGeorgi Gerganov <redacted>
Sat, 20 Sep 2025 10:33:50 +0000 (13:33 +0300)
commita3e9f2331bf9a0437af7f72c56a03eccf9521468
treedf6015e9a0656c312b4375b6bc537404e88152b3
parentfee0ce78e412655b771b2867e25e92397810806b
vulkan: Use larger loads in scalar/coopmat1 matmul (llama/15729)

I think glslang will translate an access like x[i][1].z to
OpAccessChain ... x, i, 1, 2
OpLoad float16_t ...

rather than loading all of x[i] in a single OpLoad. Change the
code to explicitly load the vector/matrix.
src/ggml-vulkan/vulkan-shaders/mul_mm.comp
src/ggml-vulkan/vulkan-shaders/types.comp
src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp