llama.cpp

CUDA: fix tensor core logic for Pascal and HIP

#4682

Merged