851062 - RIABench "2D-Test" drawing tests is 2x slower than Chrome on Windows

Status: NEW

(Core :: Graphics, defect)

Description

[Boris Zbarsky [:bzbarsky]]

Filing under graphics, because on Mac we kick their butt on this one, so it's not likely to be JS or DOM.

STEPS TO REPRODUCE:

1)  Load http://www.timo-ernst.net/misc/riabench-start/2dtest/javascript
2)  Wait for the FPS number.

EXPECTED RESULTS:  Large number (e.g. on Mac I get 42).

ACTUAL RESULTS: Small number, apparently, according to http://media.bestofmicro.com/A/F/376215/original/RIABenchWin8.png (they got 11, while the Chrome score they see is comparable to the one I see on Mac).

MORE DETAILS: See http://www.tomshardware.com/reviews/web-browser-chrome-25-firefox-19,3459-6.html

Comment 1

[Chris Peterson [:cpeterson]]

Is this still an issue? Are the RIABench drawing tests still slower on Windows Firefox than Chrome?

Comment 2

[Chris Peterson [:cpeterson]]

In the June 2013 WBGP, Firefox is still 2x slower than Chrome on the "2D-Test" drawing test:

http://www.tomshardware.com/reviews/chrome-27-firefox-21-opera-next,3534-6.html

http://media.bestofmicro.com/L/A/390862/original/RIAJS.png

Comment 3

[Guilherme Lima]

I'm on Windows 7. The problem is with hardware acceleration. Without it I got an average of 15 fps. With I got 1 fps. Tested on Firefox 26 Beta. Chrome 31 got 10 fps.

Comment 4

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Looks like lots of small clearRects, setting of fillStyle, and fillRects. Not sure why we're slow.

Comment 5

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Bas, is there any particular reason why D2D should be slow with a lot of small fillRects with different colors?

Comment 6

[Bas Schouten (:bas.schouten)]

(In reply to Robert O'Callahan (:roc) (Mozilla Corporation) from comment #5)
> Bas, is there any particular reason why D2D should be slow with a lot of
> small fillRects with different colors?

It's a pretty bad scenario for Direct2D. We're filling very few pixels per drawing call, and D2D is relatively high overhead per draw call. It'll do a lot of complicated work to do this. Fwiw, there's probably some complicated work we can do to improve the situation. But this is a pretty silly test and whatever we do would be pretty focused on this test.

Comment 7

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Actually I don't think this test is all that silly. Lots of small fills is a completely reasonable thing to do.