Javascript Performance Thread

[eltomito]

Should have known that OSX Firefox would be faster. Everything they do is OSX first.

Actually, it looks more like Ubuntu is first. Yes!

[Ver Greeneyes]

It's true that a lot of developers use OSX for what I assume are historical reasons (the percentage has always surprised me), and other use Linux (Nicolas B. Pierron uses and develops his own distribution, no less). None of the full-time SpiderMonkey devs use Windows regularly that I know of, and most don't have quick access to a VM. When h4writer was first setting up the Windows machines on AWFY he did look into the performance disparity a bit - some of it is from using the browser, which has a lot more stuff going on in the background than the shell. But some of it is from Windows being slow, and perhaps Visual Studio's generated code being slow. I don't think Hannes ever fully pinned down the cause of the remaining difference.

[Grantius]

I use Linux at home and Win7 at work, yet I find the opposite. Windows 7 with my crappy core 2 duo makes FF fly compared to my Linux one. Yes, HWA is enabled on Linux

[dao-g]

Yes, HWA is enabled on Linux

Maybe that's part of the problem...

[Omega X]

It's true that a lot of developers use OSX for what I assume are historical reasons (the percentage has always surprised me), and other use Linux (Nicolas B. Pierron uses and develops his own distribution, no less). None of the full-time SpiderMonkey devs use Windows regularly that I know of, and most don't have quick access to a VM. When h4writer was first setting up the Windows machines on AWFY he did look into the performance disparity a bit - some of it is from using the browser, which has a lot more stuff going on in the background than the shell. But some of it is from Windows being slow, and perhaps Visual Studio's generated code being slow. I don't think Hannes ever fully pinned down the cause of the remaining difference.

Regardless, its an issue. The majority of Firefox users are on Windows. Every moment that Firefox is slower on Windows, the more people that get the impression that Firefox is slow and migrate to something else. This is why e10s is such a big deal.

[Ver Greeneyes]

Oh, I agree. Though to be fair, several of the people working on hardware acceleration (e.g. Bas Schouten) do use Windows. Even though there are clearly differences in JS performance between platforms (at least on these benchmarks), I think hardware acceleration is more likely to impact perceived performance so long as things aren't going entirely off the rails (e.g. if there is a bug in the GC I would expect it to affect all platforms).

[Josa]

Seems like a dev is reading this thread (beside Dao - or maybe he talked to Hannes)... Safari is now purple and there is Chrome on Windows and on OS X. You can see that Chrome on Windows is slower then on OS X, just like Firefox. Maybe Windows is slower allocating memory?

Edit: https://bugzilla.mozilla.org/show_bug.cgi?id=1028242 - [meta] Improve octane score on windows

[Timvde]

Microsoft will open source Chakra!

I'm very curious. Jandem said he might write a blog about it.

[Peja Stija]

Oh well, half Google, half Microsoft, i've seen worse things, although I have to admit, very few.

[mayankleoboy1]

sadly no response on https://bugzilla.mozilla.org/show_bug.cgi?id=1162272 / https://github.com/chromium/octane/issues/29

[Ver Greeneyes]

On the plus side, once they open their Git repository, we'll know how they're doing it.

[Ver Greeneyes]

Wanted to mention that Jan de Mooij has been working on reducing the performance penalty of making JITcode only executable or writable, in hopes of turning it on by default. This would be a nice security enhancement, by preventing exploits from scribbling all over JITcode (and then having the JITs execute their code). It might also help catch crashes related to memory corruption by preventing the same thing. You can see the current progress on AWFY on the "Firefox (non writable jitcode)" line. There's more work to be done, but they might decide to take a regression if it's small enough, since the security advantages are so nice. The work is being tracked in bug 1215479 and its dependencies.

This work started gaining traction again after Mike Hommey (:glandium) fixed it up with Jan's help for iOS support. Then OpenBSD and HardenedBSD expressed an interest in using it, as they don't mind the performance hit (which is larger on the version they're starting out with than what you see on AWFY now). Having it on by default would be very nice, however - especially for things like Tor.

[Josa]

Branch pruning just landed. Lots of regression and lots of improvements. If all the regressions are fixed without affecting the wins, it's gonna be awesome.

[Ver Greeneyes]

It was backed out for the regressions it caused, but I agree: runtime PGO is a very exciting enhancement! I wonder if there's anything other than branch pruning they can do with it.

[koboltzz]

Sorry to ask, but where do you get your informations from about latest changes (summed up) in nightlys?
From http://statusupdates.dev.mozaws.net/ ?

And another question... they are doing some things to touch recognition code. In latest nightly touch is broken.
The 'dom.w3c_touch_events.enabled=2' do not work anymore. Anyone know what to do, to get touch input back?