llama.cpp

≈65% speedup of the AVX-512 implementation of `ggml_vec_dot_q4_0()`

#933

Merged