984961 - CSS animation consumes a significant amount of power at 163.com

Status: NEW

(Core :: Layout, defect)

Description

[Roberto Agostino Vitillo (:rvitillo)]

Attachment: Testcase 1

Idling on the 163.com homepage seems to consume significantly more energy than on IE and Chrome, particularly in terms of GPU power.

The culprit is the animated image shown in the testcase.

Comment 3

[Nicholas Nethercote [inactive]]

The animation in the attachment is still present on the front page of 163.com. I measured the attachment on Mac.

Nightly:

>     total W = _pkg_ (cores + _gpu_ + other) + _ram_ W
> #01 15.53 W = 12.00 ( 0.84 +  4.07 +  7.08) +  3.52 W
> 
> 1 sample taken over a period of 30.000 seconds
> 
> Name                               ID     CPU ms/s  User%  Deadlines (<2 ms, 2-5 ms)  Wakeups (Intr, Pkg idle)  GPU ms/s
> com.apple.Terminal                 109    271.77                                      73.43   53.34             421.16
>   firefox                          16774  135.54    71.63  0.07    0.03               69.54   51.15             421.21
>   plugin-container                 16775  131.89    83.40  1.10    0.30               3.40    1.93              0.00
>   Terminal                         208    5.58      86.01  0.00    0.00               0.43    0.27              0.00

Safari:

>     total W = _pkg_ (cores + _gpu_ + other) + _ram_ W
> #01  6.65 W =  4.74 ( 0.16 +  0.33 +  4.25) +  1.91 W
> 
> 1 sample taken over a period of 30.000 seconds
> 
> Name                               ID     CPU ms/s  User%  Deadlines (<2 ms, 2-5 ms)  Wakeups (Intr, Pkg idle)  GPU ms/s
> com.apple.Safari                   397    41.01                                       38.58   34.29             1.81    
>   com.apple.WebKit.WebContent      16699  40.93     75.42  0.00    0.00               38.42   34.12             1.81    
>   com.apple.WebKit.WebContent      16701  0.12      64.07  0.00    0.00               0.17    0.17              0.00    

Chrome:

>     total W = _pkg_ (cores + _gpu_ + other) + _ram_ W
> #01 12.50 W =  9.56 ( 0.85 +  2.59 +  6.12) +  2.94 W
> 
> 1 sample taken over a period of 30.000 seconds
> 
> Name                               ID     CPU ms/s  User%  Deadlines (<2 ms, 2-5 ms)  Wakeups (Intr, Pkg idle)  GPU ms/s
> com.google.Chrome                  398    473.34                                      280.61  186.95            99.76   
>   Google Chrome Helper             16736  181.79    68.18  0.00    0.00               158.78  118.16            99.77   
>   Google Chrome Helper             16741  161.92    74.78  0.00    0.00               60.48   33.92             0.00    
>   Google Chrome                    16732  131.60    70.08  0.27    0.17               61.38   34.89             0.00    

Firefox is hammering the GPU in this case.

Comment 4

[Benoit Girard (:BenWa)]

This is where bug 1124325 comes in handy. Can we get the same numbers for a simpler test:
http://people.mozilla.org/~nfroyd/power-test-pages/colorflash.html

I want to know here if the difference is caused by partial present or we're just having a worse layer tree.

Comment 5

[Benoit Girard (:BenWa)]

Also Safari is getting less than 60 wake-ups which seems suspicious. I wonder if they're even running at 60 FPS like we are. Are we getting ~3.4 wakeups in our content process? Maybe we're not going to sleep frames? I'm not sure the values look correct.

Also keep in mind that on OSX you need to measure the WindowServer activity as well. Particularly since Safari offloads a lot of the computation there then Firefox and Chrome.