By some feedback of AI compiler performance tuner, the compiler based on XOC is 20% average faster than LLVM based.