llama.cpp

vulkan: only use M-sized matmul on Apple GPUs

#5412

Merged