Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

The approach to these vectorized parsers is generally to find some part of the parsing task that doesn't have a dependency chain through every dang byte, like a naively written parser would, then do that first, then do some harder stuff after. I'm pretty used to it now so I lack perspective on whether this is weird, but in general finding long dependency chains and getting rid of them or replacing them with several shorter dependency chains will make software a lot faster.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: