Bug 120012 – 100% cpu usage for "transition: opacity" animation

• Status: CLOSED
• Resolution: FIXED
• Component: WebKit2

[Luciano Wolf]

Created attachment 209095 [details]
Proposed fix - WIP - doesn't include ChangeLog yet.

While running leaves demo, the CPU tops at 100% for WebProcess. Even simple examples like LayoutTests/compositing/repaint/opacity-between-absolute.html does the same.

Doing some research I've found that CoordinatedLayerTreeHost::scheduleAnimation() does set a timer for animation and right after that it schedules again the same timer with 0. This behavior could be avoided if we call scheduleLayerFlush() before m_layerFlushTimer.startOneShot(m_coordinator->nextAnimationServiceTime()).

I've detected this issue on NIX port, but I wasn't able to compile Qt, GTK or EFL.

[JeongBong Seo]

It also detected and resolved by your idea with Qt port.

[Luciano Wolf]

So, is this patch useful? Should I updated it with ChangeLog info and ask for review?

[Noam Rosenthal]

(In reply to comment #2)
> So, is this patch useful? Should I updated it with ChangeLog info and ask for review?

Yes please!

[Luciano Wolf]

Created attachment 209793 [details]
Patch

[JeongBong Seo]

I added a reivew as a comment. (Because, I don't have the auth.)

Source/WebKit2/WebProcess/WebPage/CoordinatedGraphics/CoordinatedLayerTreeHost.cpp:399
>      scheduleLayerFlush();

I think, scheduleLayerFlush() at this moment is meaningless.
How about erase that and insert below?

if(!m_layerFlushSchedulingEnabled)
    return;

[WebKit Commit Bot]

Comment on attachment 209793 [details]
Patch

Clearing flags on attachment: 209793

Committed r154711: <http://trac.webkit.org/changeset/154711>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Luciano Wolf]

(In reply to comment #5)
> I added a reivew as a comment. (Because, I don't have the auth.)
> 
> Source/WebKit2/WebProcess/WebPage/CoordinatedGraphics/CoordinatedLayerTreeHost.cpp:399
> >      scheduleLayerFlush();
> 
> I think, scheduleLayerFlush() at this moment is meaningless.
> How about erase that and insert below?
> 
> if(!m_layerFlushSchedulingEnabled)
>     return;

Sorry, I missed your comment among commit messages :)

I'll test and provide an updated patch. Thanks for your suggestion.