I don't think anyone has specifically investigated this, but here's three random notes:
(1) We optimize hardest for x86, since our largest platform is Windows. Since JSC's largest platform is OS X, they probably optimize for x64. (2) JM's design does not map well to our x64 boxing format. IonMonkey does a lot better and has room to grow. See the "Regression Testing" test view for IonMonkey lines - the gaps between JM there are wider. (3) JSC has a different x64 boxing scheme that favors objects, whereas ours favors doubles. That could help explain the gaps being very different on x86/x64 Kraken (which is double heavy) versus x86/x64 v8bench (which is object heavy). While neither of our engines have a big score difference on Kraken on x86 versus x64, both are much slower on v8bench running on x64 versus running on x86. (V8 is slower, too, but not as much.) -David On 08/22/2012 12:53 PM, Tom Schuster wrote: > We are pretty close here http://arewefastyet.com/?a=b&machine=11 on > v8, but we get finished right here > http://arewefastyet.com/?a=b&machine=12. > So I am wondering if somebody investigated the reason for this fairly large > gap? > > Cheers, > Tom > _______________________________________________ > dev-tech-js-engine-internals mailing list > [email protected] > https://lists.mozilla.org/listinfo/dev-tech-js-engine-internals > _______________________________________________ dev-tech-js-engine-internals mailing list [email protected] https://lists.mozilla.org/listinfo/dev-tech-js-engine-internals
