958823 - Three.js canvas renderer is super-slow in Firefox compared to Chrome on Android

Status: RESOLVED INACTIVE

(Core :: Graphics: Canvas2D, defect)

Description

[Chris Lord [:cwiiis]]

Going to the given URL in Firefox for Android results in 0-2fps, where as it's a smooth 30+fps in Chrome. I'm guessing that for whatever reason, this isn't hitting our accelerated canvas fast-paths?

Tested on a Google Nexus 5 with today's Nightly.

Comment 1

[James Willcox (:snorp) ([e-mail]) (he/him)]

Yeah I noticed this too. Weird. I'll look at it.

Comment 2

[Mayank Bansal]

FWIW, this demo is slow in windows7 also, compared to chrome.

Nightly+omtc+new textures : 23FPS
Chrome32 : 43 fps

Comment 3

[Milan Sreckovic [:milan] (needinfo for best results)]

On the Mac: nightly ~ 3-7fps, Chrome: 60fps.  Do we know if this is canvas or webgl side that's slowing things down?

Comment 4

[Milan Sreckovic [:milan] (needinfo for best results)]

Actually, I think I can answer that - it's Canvas2D.  WebGL seems to be OK on it's own (e.g., ~60fps).

Comment 5

[Milan Sreckovic [:milan] (needinfo for best results)]

On the Mac, Fill and FillRect add up to 60%.

Comment 6

[James Willcox (:snorp) ([e-mail]) (he/him)]

(In reply to Milan Sreckovic [:milan] from comment #5)
> On the Mac, Fill and FillRect add up to 60%.

Yeah, looks like Skia is doing a readback here for some reason. Digging some more.

Comment 7

[James Willcox (:snorp) ([e-mail]) (he/him)]

Looks like it's drawing one canvas into another, which is known to be slow right now.

Comment 8

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Gets fast with shared-GLContexts!

Comment 9

[James Willcox (:snorp) ([e-mail]) (he/him)]

(In reply to Vladimir Vukicevic [:vlad] [:vladv] from comment #8)
> Gets fast with shared-GLContexts!

It seems that way to me too, did you try it? It makes sense.

Comment 10

[James Willcox (:snorp) ([e-mail]) (he/him)]

So https://bug935688.bugzilla.mozilla.org/attachment.cgi?id=828213 is fast for me, which does the same thing, but this bug is still present on Mac at least. Strangeness.

Comment 11

[James Willcox (:snorp) ([e-mail]) (he/him)]

(With bug 939276 patches applied, that is)

Comment 12

[James Willcox (:snorp) ([e-mail]) (he/him)]

Ok, not sure what I was doing before, but this does in fact appear to be fixed by bug 939276

Comment 13

[Alex]

Hi James, not sure what you was doing in #939276 but in Firefox 29(mac) this still creeps with 2-4 Fps.

Comment 14

[James Willcox (:snorp) ([e-mail]) (he/him)]

(In reply to Alex from comment #13)
> Hi James, not sure what you was doing in #939276 but in Firefox 29(mac) this
> still creeps with 2-4 Fps.

Ah, I was talking about the performance with SkiaGL enabled, which is not the default (yet). I don't think this is going to ever be fast under the default CoreGraphics backend, unfortunately.

Comment 15

[gokcehankara]

Not sure if it helps but, three.js example given here seems to be quite complicated. I get the slowdown even for the simplest canvas renderer example:

http://threejs.org/examples/canvas_geometry_cube.html

On my nexus 7 (android 4.4.4 - kitkat) I get the following numbers:

Firefox for android (v31.0): 3-4 fps
Chrome for android (v36.0.1985.131): 60 fps (capped)

By the way, does this mean it will have the same problem in FirefoxOS?

Comment 16

[BugBot [:suhaib / :marco/ :calixte]]

The bug assignee didn't login in Bugzilla in the last 7 months.
:lsalzman, could you have a look please?
For more information, please visit auto_nag documentation.