My current suspicion is that most of the time is spent in OSR exit computation. We know we can do better there, so it would be useful to be able to confirm this hypothesis on a variety of benchmarks.
Created attachment 206109 [details] the patch
Landed in http://trac.webkit.org/changeset/152432
> My current suspicion is that most of the time is spent in OSR exit computation. Did this turn out to be true?
(In reply to comment #3) > > My current suspicion is that most of the time is spent in OSR exit computation. > > Did this turn out to be true? Currently LLVM compile times reduce by 30% if it doesn't have to deal with OSR exit marshaling. So, no, it's not "most" of the time, right now. But if you consider that we're also spending some significant fraction of time setting up pass managers, contexts, modules, execution engines, etc., I suspect that it'll be more than 30% if those other issues are fixed.