[Webkit-unassigned] [Bug 31415] [v8] do not copy data twice when converting short v8 string into WebCore::String

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=31415





--- Comment #15 from anton muhin <antonm at chromium.org>  2009-12-01 09:52:12 PST ---
(In reply to comment #13)
> (From update of attachment 43872 [details])
> Darin Fisher and I discussed it a bit. Although he's noted the style issues, he
> didn't mean to say the discussion we had before is concluded.
> 
> Usually, perf improvement should measurably improve perf, so if there is no
> test but you say it can easily be created then lets add a test, measure and
> know what we are gaining. Having a test also prevents possible regressions. We
> have a v8-specific set of tests in src/data/dom_perf - could you add a specific
> test here?
> 
> Again, if there is no measurable difference, I don't see why make the change.
> Agree there could be an extra copy of up to 16 bytes to a stack array. But if
> it's impossible to measure, why split the code at all?
> 
> r- because it needs a test.

For Dromaeo DOM core (which is one of our primary benchmarks now and the only
benchmark we're lagging Safari for now), that gives ~2.5% speedup for overall
score (4 runs, dropping the slowest) and, more importantly, ~10% for DOM query
(which is, together with DOM Modification) is there we're behind Safari.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.