We should be able to use the code, cmake files, etc. from llama.cpp as is:

#471