1059557 - css animated throbber causes heavy cpu load

Status: RESOLVED FIXED

(Firefox :: Theme, defect)

Description

[Jeff Muizelaar [:jrmuizel]]

We have much higher cpu load from this animation with limited benefit. This animation seems to cause high cpu load even when the throbber is not visible.

Comment 1

[Dão Gottwald [:dao]]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #0)
> We have much higher cpu load from this animation with limited benefit. This
> animation seems to cause high cpu load even when the throbber is not visible.

This would be bug 962594, which we worked around in bug 1016942. So this shouldn't be happening anymore...

Comment 2

[:Gavin Sharp [email: [e-mail]]]

Jeff, is comment 1 wrong?

Comment 3

[Jeff Muizelaar [:jrmuizel]]

(In reply to :Gavin Sharp [email: [e-mail]] from comment #2)
> Jeff, is comment 1 wrong?

Yes. I still have high load when non-visible tabs have a spinning throbber. The load is also especially high when the tabs are still visible? Is there a performance comparison somewhere to the previous animated png throbber? I see 39% Nightly cpu usage and 46% WindowServer usage with a visible throbber. Is this expected?

Comment 4

[Dão Gottwald [:dao]]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #3)
> (In reply to :Gavin Sharp [email: [e-mail]] from comment #2)
> > Jeff, is comment 1 wrong?
> 
> Yes. I still have high load when non-visible tabs have a spinning throbber.

You mean tabs that are scrolled off-screen? Would the png throbber stop animating and straining the CPU in that case?

> The load is also especially high when the tabs are still visible? Is there a
> performance comparison somewhere to the previous animated png throbber? I
> see 39% Nightly cpu usage and 46% WindowServer usage with a visible
> throbber. Is this expected?

That's to some extent expected, as the png throbber has a lower frame rate. The CSS animation is expected to be smoother. Another expectation is that off-main-thread animations will alleviate the performance impact.

Comment 5

[Jeff Muizelaar [:jrmuizel]]

(In reply to Dão Gottwald [:dao] from comment #4)
> (In reply to Jeff Muizelaar [:jrmuizel] from comment #3)
> > (In reply to :Gavin Sharp [email: [e-mail]] from comment #2)
> > > Jeff, is comment 1 wrong?
> > 
> > Yes. I still have high load when non-visible tabs have a spinning throbber.
> 
> You mean tabs that are scrolled off-screen? Would the png throbber stop
> animating and straining the CPU in that case?

I'm not sure. I never noticed cpu load from the png throbbers though.

> > The load is also especially high when the tabs are still visible? Is there a
> > performance comparison somewhere to the previous animated png throbber? I
> > see 39% Nightly cpu usage and 46% WindowServer usage with a visible
> > throbber. Is this expected?
> 
> That's to some extent expected, as the png throbber has a lower frame rate.
> The CSS animation is expected to be smoother. Another expectation is that
> off-main-thread animations will alleviate the performance impact.

We don't have off-main-thread animations on right now. I expect it might be worth waiting till then before enabling the css-animation.

Comment 6

[Dão Gottwald [:dao]]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #5)
> We don't have off-main-thread animations on right now. I expect it might be
> worth waiting till then before enabling the css-animation.

We've been backing out the CSS throbber from Firefox 32 and 33 and we should probably keep doing that until off-main-thread animations are enabled.

Comment 7

[Dão Gottwald [:dao]]

[Tracking Requested - why for this release]: see comment 6

Comment 8

[Lawrence Mandel [:lmandel] (use needinfo)]

(In reply to Dão Gottwald [:dao] from comment #6)
> We've been backing out the CSS throbber from Firefox 32 and 33 and we should
> probably keep doing that until off-main-thread animations are enabled.

+1 Can we back out the CSS throbber altogether and reland the it after OMTC animations land?

Comment 9

[Jared Wein [:jaws] (please needinfo? me)]

Yeah, I agree, we should just back out. It doesn't appear that OMTC will be enabled on Windows Nightly for quite a few releases still (please correct me if I'm wrong).

Comment 10

[Jeff Muizelaar [:jrmuizel]]

(In reply to Jared Wein [:jaws] (please needinfo? me) from comment #9)
> Yeah, I agree, we should just back out. It doesn't appear that OMTC will be
> enabled on Windows Nightly for quite a few releases still (please correct me
> if I'm wrong).

This needs Off Main Thread Animations not just OMTC. OMTC will hopefully be in the upcoming release.

Comment 11

[Jared Wein [:jaws] (please needinfo? me)]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #10)
> (In reply to Jared Wein [:jaws] (please needinfo? me) from comment #9)
> > Yeah, I agree, we should just back out. It doesn't appear that OMTC will be
> > enabled on Windows Nightly for quite a few releases still (please correct me
> > if I'm wrong).
> 
> This needs Off Main Thread Animations not just OMTC. OMTC will hopefully be
> in the upcoming release.

Sorry and thanks for the clarification.

Comment 12

[Dão Gottwald [:dao]]

https://hg.mozilla.org/integration/fx-team/rev/4af29b25a7bf

Comment 13

[Tim Nguyen :ntim]

https://hg.mozilla.org/mozilla-central/rev/4af29b25a7bf

Comment 14

[Lawrence Mandel [:lmandel] (use needinfo)]

jaws - 34 is marked as affected. Can you please submit an Aurora uplift request?

Comment 15

[Lawrence Mandel [:lmandel] (use needinfo)]

Erm, rather, can you back this out of 34 as I see it has been backed out of 32, 33, and 35?

Comment 16

[Dão Gottwald [:dao]]

This was backed out in bug 1016434.

Comment 17

[Liz Henry (:lizzard) (relman/hg->git project)]

This was verified in bug 1016434.