What would you like you use instead? C/C++ code which targets generic x86_64 is a few percent slower than march=native but that still leaves it way faster than most other languages.