1070722 - [Mac] Imgur slow to scroll. Single image display item takes 1.2 seconds to draw

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Benoit Girard (:BenWa)]

URL: http://imgur.com/a/U8pSW

http://people.mozilla.org/~bgirard/cleopatra/#report=49e182f4df7923395dd0f69f9f58a3299ed40bf0

I've noticed imgur being really bad lately. It feels almost unusable without adblock.

Chrome can scroll this page smoothly. Comparing the two, their progressive decode is much more noticeable then ours. But if you look at the profile the problem is the time it takes to draw the image.

Jeff do you remember what the resample_byte_*_cpp_vector paths are? Is this high quality downscale?

Comment 1

[Benoit Girard (:BenWa)]

[Tracking Requested - why for this release]:
This is a severe performance regression.

A regression window here would really help.

Comment 2

[Jeff Muizelaar [:jrmuizel]]

The profile is spending time in hq downscale. Perhaps this is broken caching of downscaled results.

Comment 3

[Tim Taubert [:ttaubert] (inactive)]

This is hopefully the right range:

Last good nightly: 2014-09-03
First bad nightly: 2014-09-04

Pushlog:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=e58842c764dd&tochange=776fa9cf70cd

Comment 4

[Markus Stange [:mstange]]

Tiling (bug 982338) is in that range.

Maybe we're doing a whole-image downscale for each tile that an image covers?

Comment 5

[Tim Taubert [:ttaubert] (inactive)]

Yup, that's it. Confirmed through bisecting.

Comment 6

[Matt Woodrow (:mattwoodrow)]

Attachment: Disable Quartz high-quality downscaling

I disabled the OptimizeSourceSurface code since we no longer get much benefit from retaining the CGImage and it lets us skip a copy of the image data.

Comment 7

[Timothy Nikkel (:tnikkel)]

Comment on attachment 8494051 [details] [diff] [review]
Disable Quartz high-quality downscaling

And enable our own HQ downscaler in the prefs on mac?

Comment 8

[Matt Woodrow (:mattwoodrow)]

Attachment: Disable Quartz high-quality downscaling v2

Comment 9

[Timothy Nikkel (:tnikkel)]

I guess we won't need this code anymore?
http://mxr.mozilla.org/mozilla-central/source/image/src/Decoder.cpp?force=1#255

Comment 10

[Jeff Muizelaar [:jrmuizel]]

Comment on attachment 8494061 [details] [diff] [review]
Disable Quartz high-quality downscaling v2

Review of attachment 8494061 [details] [diff] [review]:
-----------------------------------------------------------------

::: gfx/2d/DrawTargetCG.cpp
@@ +249,5 @@
>  
>  TemporaryRef<SourceSurface>
>  DrawTargetCG::OptimizeSourceSurface(SourceSurface *aSurface) const
>  {
> +  return aSurface;

Can we do this change separately?

Comment 11

[Matt Woodrow (:mattwoodrow)]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #10)

> 
> Can we do this change separately?

Sure, how separately do you want? Different patches, different landings, different bug?

Comment 12

[Jeff Muizelaar [:jrmuizel]]

Separated landings, so that regression windows find the other one.

Comment 13

[Matt Woodrow (:mattwoodrow)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/b5705fbed44f

Landed the first half, will land the rest soon.

Comment 14

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

I had to back this out in https://hg.mozilla.org/integration/mozilla-inbound/rev/a0838b04a1e4 for reftest failures: https://tbpl.mozilla.org/php/getParsedLog.php?id=48996619&tree=Mozilla-Inbound

Comment 15

[Matt Woodrow (:mattwoodrow)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/6fb508e952b1

Comment 16

[Matt Woodrow (:mattwoodrow)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/a4ae53da4fe6

Comment 17

[Matt Woodrow (:mattwoodrow)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/4319ae3a5d2f

Comment 18

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/6fb508e952b1
https://hg.mozilla.org/mozilla-central/rev/a4ae53da4fe6
https://hg.mozilla.org/mozilla-central/rev/4319ae3a5d2f

Comment 19

[Benoit Girard (:BenWa)]

I justed tested with today' night. Things are better. We're still a tad jerky scrolling pages with large images however.

Comment 20

[Matt Woodrow (:mattwoodrow)]

You should get a profile :)

I've noticed huge janks when the flash ad becomes visible, you wouldn't be getting that if you have adblock though.

Comment 21

[Benoit Girard (:BenWa)]

You want to solve this here or spin off a new bug?

Comment 22

[Matt Woodrow (:mattwoodrow)]

Probably a new bug, we've already fixed something here.

Comment 23

[Matt Woodrow (:mattwoodrow)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/0f14e8b69354

Comment 24

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/0f14e8b69354