llama.cpp

aa014d7e - Use mutex instead of atomics for vk_instance counters