567675 - Important performance problem on SVG Images with transluscency

Status: RESOLVED WORKSFORME

(Core :: SVG, defect)

Description

[Vincent Hardy]

User-Agent:       Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.6; en-US; rv:1.9.3a4webm) Gecko/20100518 MozillaDeveloperPreview/3.7a4webm
Build Identifier: Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.6; en-US; rv:1.9.3a4webm) Gecko/20100518 MozillaDeveloperPreview/3.7a4webm

The SVG demo shown in the above URL relies on animation of the opacity of small SVG images. In particular, there is a kind of fish-eye effect when the user hovers on the images where images are scaled, rotated and their opacity is modified. In Firefox, the performance is degraded greatly.

Reproducible: Always

Steps to Reproduce:
Look at the video at: http://svg-wow.org/photoAlbum/light-table.html to see the demo interactivity and expected performance (the video was recorded using WebKit to run the content):

1. Open http://svg-wow.org/photoAlbum/light-table.svg
2. Wait for the images to download (they trickle down from the top of the screen as they download)
3. Move the cursor over the images after they have all loaded.
Actual Results:  
The performance is poor.

Expected Results:  
Faster compositing of images with opacity.

Comment 1

[Mardeg]

By "animation of the opacity" did you mean the CSS opacity style, or the opacity attribute in SVG, or SMIL animation of a <filter> applying opacity?
Please attach a reduced testcase that doesn't rely on 14 external scripts.

Comment 2

[Daniel Holbert [:dholbert]]

I'm guessing this is bug 562815, if it's due to SMIL animation on the opacity attribute or property (which are mostly handled the same, and both suffer from bug 562815).  If this is a <filter>, on the other hand, I'm not as sure what the root cause would be.

Anyway, yes -- reduced-testcase-wanted, but I'm guessing this is bug 562815.

Comment 3

[(mostly gone) XtC4UaLL [:xtc4uall]]

Tested http://svg-wow.org/photoAlbum/light-table.svg against Mozilla/5.0 (Windows NT 5.1; rv:9.0a1) Gecko/20110823 Firefox/9.0a1 ID:20110823085420 (http://hg.mozilla.org/mozilla-central/rev/8f2530ae725a) what contains the Fix for Bug 562815  I notice Improvement, but still not on par with Chrome/15.0.860.0.

Vincent, can you provide a reduced Testcase as requested to be able to fix the outstanding Issues?

Comment 4

[Brian Birtles (:birtles)]

The testcase also includes numerous animations on transforms. I'm seeing a marked improvement on the Santa's Workshop demo with the SVG transform patches in bug 602759. Those patches are nearing the end of the review stage (hopefully) so let's see what difference those make. I be able to spin a try server build later today if I finish getting through all the review feedback.

Comment 5

[Vincent Hardy]

(In reply to XtC4UaLL [:xtc4uall] from comment #3)
> Tested http://svg-wow.org/photoAlbum/light-table.svg against Mozilla/5.0
> (Windows NT 5.1; rv:9.0a1) Gecko/20110823 Firefox/9.0a1 ID:20110823085420
> (http://hg.mozilla.org/mozilla-central/rev/8f2530ae725a) what contains the
> Fix for Bug 562815  I notice Improvement, but still not on par with
> Chrome/15.0.860.0.
> 
> Vincent, can you provide a reduced Testcase as requested to be able to fix
> the outstanding Issues?

Unfortunately no, I wont be able to work on a test case any time soon. Sorry.

Comment 6

[Ryan VanderMeulen [:RyanVM]]

Definitely slower than Chrome. Win7 D2D here.

Comment 7

[(dormant account)]

This is a valid SVG bug, but I don't think this Snappy-level responsiveness.

Comment 8

[Boris Zbarsky [:bzbarsky]]

Is this still a problem?  Performance is instantaneous for me (on Mac, though).

Comment 9

[Vincent Hardy]

I still see a big performance gap between Firefox and WebKit (Chrome or Safari). Testing on a Mac Book Pro, OS X 10.8.2

Comment 10

[Boris Zbarsky [:bzbarsky]]

OK.  I'd like to understand your steps to reproduce, then.  Because as I said, the updates are instantaneous for me, likewise on MacBook Pro, OS X 10.8.2.

Comment 11

[(mostly gone) XtC4UaLL [:xtc4uall]]

On Win 7/D2D, no Change to Comment 3.
Profile against Mozilla/5.0 (Windows NT 6.1; WOW64; rv:23.0) Gecko/20130402 Firefox/23.0 ID:20130402030843 CSet: aae004a3c5d9 with moving the Mouse Pointer over the Images several Times.

http://people.mozilla.com/~bgirard/cleopatra/#report=1f53939750c80ea68b41cef33beaa6abb2c652fc

Comment 12

[Boris Zbarsky [:bzbarsky]]

Nothing in that profile is showing particular slowness, fwiw.  About 1/3 is painting, 1/3 is processing the mouse events in the JS, and 1/3 is just idle...

Comment 13

[Vincent Hardy]

To Boris' question: The steps are just as described in the original bug description. I just tested Firefox 20.0 and the performance is visibly slower than Chrome and Safari.

Comment 14

[Boris Zbarsky [:bzbarsky]]

I just can't reproduce that.  The behavior looks identical for me in Chrome, Safari, and current Gecko.  :(

Comment 15

[Brian Birtles (:birtles)]

On my Win7 machine Firefox 21 has a visibly lower frame rate than Chrome 26. Current nightly is noticeably faster than Firefox 21 but still somewhat less smooth than Chrome. Vincent, what performance do you see on current nightly?

Comment 16

[Vincent Hardy]

Hi Brian, good to hear from you! :-)

Yes, on the Mac too the performance on the nightly seems quite a bit better. I had turned off opacity on the demo a while back, but with regular opaque images it seems the problem is fixed in nightlies. Way better performance than on Firefox 20 (Mac).
Thanks!

Comment 17

[Robert Longson [:longsonr]]

I guess we can close this as WFM given that then.