llama.cpp

f09b7cb6 - rm get_work_group_size() by local cache for performance (#8286)