On 3/7/16 4:42 PM, 罗勇刚(Yonggang Luo) wrote:
My conclution: we really need improve the performance of Object,
Please file a bug on JS engine.
That said, some of your results are a bit suspect. For example, the
"testObjectSingle" number for Chrome shows that each set takes 4.9ns on
your hardware. On mine, it takes 0.7ns (my number there is 7ms, not
49ms). 0.7ns is about how long that test takes if I comment out the
loop body entirely. So what that test is really measuring is whether
the JIT manages to figure out that the earlier sets are dominated by
later ones and optimize them all out entirely. That's a good thing to
be able to do, but has nothing to do with the performance of property
sets per se.
For what it's also worth, the Array test runs faster than in Chrome (by
almost 3x) for me on Mac, while the uint32Array one runs slower by 3x...
The testObject test is definitely worth looking into, though I expect
it's caused by bug 1091978: profile shows we keep missing in the
SetProperty IC and having to do a vmcall.
-Boris
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform
