[Webkit-unassigned] [Bug 64582] JSC JIT does not inline GC allocation fast paths

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=64582





--- Comment #7 from Filip Pizlo <fpizlo at apple.com>  2011-07-14 21:55:36 PST ---
(In reply to comment #6)
> (In reply to comment #1)
> > Created an attachment (id=100924)
 --> (https://bugs.webkit.org/attachment.cgi?id=100924&action=review) [details] [details]
> > the patch
> > 
> > This patch works gives a 1.3% speed-up on V8, and is neutral on SunSpider.
> > 
> > I've lived on it for a day or so, and it seems to work.  It passes all tests.  But I'm not sure if I'm handling global objects (for initializing the structure in op_new_object) correctly, so I'd appreciate some feedback on that!  Separately I tested the speed-up of op_new_object in isolation, and it wasn't much (statistically significant but fraction of a percent); it definitely appears that op_create_this is doing most of the work.
> 
> I'm kind of surprised it doesn't help a microbenchmark of op_new_object, or do you mean only inlining op_new_object doesn't seem to help any of the benchmarks we track? (which is less surprising)

To clarify: I only meant that it didn't help V8 or SunSpider by more than a fraction of a percent.  I haven't tried a micro benchmark; I "declared victory" once I saw that it gave a slight win on V8.

It definitely helps micro benchmarks (see below).

mytest2.js:
for (var i = 0; i < 10000000; ++i) {
    var x = new Object();
}

With the patch:
[pizlo at minime OpenSource] time WebKitBuild/Release/jsc mytest2.js 

real    0m0.277s
user    0m0.270s
sys    0m0.006s

Without the patch:
[pizlo at minime OpenSource] time ../../secondary/OpenSource/WebKitBuild/Release/jsc mytest2.js 

real    0m0.474s
user    0m0.264s
sys    0m0.008s

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