Better than what? And do you use `-mavx2` or do you let it target baseline x86_64 and miss out on 8-float vectors? How do you make sure its autovectorisation is

Better than what? And do you use `-mavx2` or do you let it target baseline x86_64 and miss out on 8-float vectors? How do you make sure its autovectorisation is successful?