llama.cpp

vulkan: optimize coopmat2 q4_k/q5_k dequant functions.

#11206

Merged