llama.cpp

CUDA: fix matrix multiplication algorithm choice

#8102

Merged