llama.cpp

CUDA: Check PTX version on host side to guard PDL dispatch

#23530

Merged