Dumb question - but what would you use to benchmark clang itself? I thought about just taking the time to run the entire lit test suite but dunno if that's a good measure or not?
@neilhenning depends on for what...
But the lit test suite isn't great at much iirc because it spends so much time on color er cases and diagnosing things -- not usually the hot path.
A big project's build works well for general performance, llvm and clang itself work reasonably well.
For more targeted things, I've typically generated a test case to exercise that pattern.
@chandlerc oh actually that was a late night doh - bootstrapping clang itself is the obvious benchmark to use thanks!