llama.cpp

[SYCL] replace get_work_group_size() by local cache for performance

#8286

Merged