llama.cpp

cuda : use std::string instead of VLA for archName

#30463

Merged