577623 - Slower performance on that site compared to other browsers

Status: RESOLVED INCOMPLETE

(Core :: General, defect)

Description

[Martijn Wargers (dead)]

That demo page is slower in Firefox4 beta1, according to a Slashdot poster, see:
http://developers.slashdot.org/comments.pl?sid=1713004&cid=32844616

This were his results:
Chrome 6: 31 FPS
Opera 10.60: 46 FPS
Safari 5.0: 25 FPS; visually poor results
Internet Explorer 9: 19 FPS
Firefox 4.0 Beta 1: 19 FPS

I don't know if hardware acceleration was turned on for him or not. He doesn't mention it, so he probably didn't have turned it on.

I get these results:
Chrome 5: 22/23fps
Firefox trunk (with hardware acceleration on windows 7): 17fps


Boris also commented in the bug:
"
In Firefox, about 25% of the time on that page is spent in anything canvas-related. Another 20% is spent painting.

Also, 10% is spent allocating arrays, 20% is spent sorting arrays. Another 15% is spent running other JavaScript of various sorts.

So the issue here is not exactly canvas performance...
"

Comment 1

[Boris Zbarsky [:bzbarsky]]

Yeah, I talked to dvander about this today.  With jaegermonkey, the array sort drops to 5%, but more time is spent in the code itself, as far as I can see...  Unfortunately, it's really hard to debug that sort of perf issue in jit-generated code.  :(  For what it's worth, the numbers I get (on Mac) are more or less identical to yours.

Martijn, do you think you can put together a self-contained version of the page just in case it goes away?

It might still be worth looking at the canvas and painting stuff here, but we'd have to get its time down to pretty much 0 to hit Chrome's numbers over here, so we'll need js engine speedups here no matter what.

Comment 2

[Jeff Muizelaar [:jrmuizel]]

There's definitely some unnecessary overhead we're paying for in the cairo to quartz translation. For example we're spending 3.3% in _cairo_quartz_setup_state. Bug 544102 and bug 544357 are ones that we have open on this this problem.

Comment 3

[Martijn Wargers (dead)]

Attachment: zipped up file, based on url (for backup)

Comment 4

[Martijn Wargers (dead)]

Ugh, in the zipped attachment that I made, I forgot the actual html to test with and the url is gone now, too, so I guess this bug is incomplete.
Sorry about hat.