llama.cpp

[SYCL] Use batched mul_mat pathway

#5591

Merged