llama.cpp

CUDA: use arch list for compatibility check

#11775

Merged