llama.cpp

vulkan: Implement split_k for coopmat2 flash attention.

#12627

Merged