llama.cpp

CUDA: fixed tensor cores not being used on RDNA3

#4697

Merged