ggml-cuda: pad matmul weight rows to avoid cache-set aliasing#78

roberteg16 merged 2 commits into

gfx11AMD-Ecosystem/llama.cpp:gfx11from

rogarcia.padd-gemm-weight-rows-to-avoid-cache-setAMD-Ecosystem/llama.cpp:rogarcia.padd-gemm-weight-rows-to-avoid-cache-setCopy head branch name to clipboard