Closed Bug 2000149 Opened 2 months ago Closed 7 days ago

Let WebRender layer compositor ride the trains to release on Windows

Categories

(Core :: Graphics: WebRender, task)

Unspecified
Windows
task

Tracking

()

RESOLVED FIXED
149 Branch
Tracking Status
relnote-firefox --- ?
firefox149 --- fixed

People

(Reporter: sotaro, Assigned: sotaro)

References

(Blocks 2 open bugs)

Details

(Keywords: perf-alert)

Attachments

(1 file)

No description provided.
Blocks: 1959009
Assignee: nobody → sotaro.ikeda.g
Pushed by sikeda.birchill@mozilla.com: https://github.com/mozilla-firefox/firefox/commit/6f26cb765790 https://hg.mozilla.org/integration/autoland/rev/6b7aa60a1fd5 Let WebRender layer compositor ride the trains to release on Windows r=gfx-reviewers,lsalzman
Status: NEW → RESOLVED
Closed: 2 months ago
Resolution: --- → FIXED
Target Milestone: --- → 147 Branch
Regressions: 2001612
Pushed by abutkovits@mozilla.com: https://github.com/mozilla-firefox/firefox/commit/cbafe46ac95f https://hg.mozilla.org/integration/autoland/rev/9f6a8d0d5d46 Revert "Bug 2000149 - Let WebRender layer compositor ride the trains to release on Windows r=gfx-reviewers,lsalzman" for causing Bug 2001612.
Status: RESOLVED → REOPENED
Flags: needinfo?(sotaro.ikeda.g)
Resolution: FIXED → ---
Target Milestone: 147 Branch → ---
Flags: needinfo?(sotaro.ikeda.g)
Depends on: 2005100
Depends on: 2005102
Depends on: 2005103

(In reply to Atila Butkovits from comment #5)

Backed out as requested: https://hg-edge.mozilla.org/integration/autoland/rev/9f6a8d0d5d46cc03884cbee564ba1c9deccdb82b

Perfherder has detected a browsertime performance change from push 9f6a8d0d5d46cc03884cbee564ba1c9deccdb82b.

No action is required from the author; this comment is provided for informational purposes only.

Regressions:

Ratio Test Platform Options Absolute values (old vs new) Performance Profiles
19% speedometer VueJS-TodoMVC/CompletingAllItems/Sync windows11-64-24h2-nightlyasrelease fission webrender 0.65 -> 0.77 Before/After
19% speedometer Preact-TodoMVC/CompletingAllItems/Sync windows11-64-24h2-nightlyasrelease fission webrender 0.82 -> 0.97 Before/After

Improvements:

Ratio Test Platform Options Absolute values (old vs new) Performance Profiles
13% speedometer Angular2-TypeScript-TodoMVC/CompletingAllItems/Async windows11-64-24h2-nightlyasrelease fission webrender 1.22 -> 1.07 Before/After
10% speedometer Elm-TodoMVC/Adding100Items/Sync windows11-64-24h2-nightlyasrelease fission webrender 29.85 -> 26.94 Before/After
10% speedometer Elm-TodoMVC/Adding100Items windows11-64-24h2-nightlyasrelease fission webrender 32.15 -> 29.08 Before/After
10% speedometer Preact-TodoMVC/Adding100Items/Sync windows11-64-24h2-nightlyasrelease fission webrender 0.71 -> 0.64 Before/After

Need Help or Information?

If you have any questions, please reach out to bacasandrei@mozilla.com. Alternatively, you can find help on Slack by joining #perf-help, and on Matrix you can find help by joining #perftest.

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests.

Keywords: perf-alert

From Bug 2001612 comment 5,


When I look at the graphs on that page it seems like a clear regression to me.

If we're going to regress we should try to have a theory of why and understand what the trade offs are.

The only thing I recall seeing about the advantages was this message from Glenn:
We have initial measurements only from the prototype (where we saw ~20% savings in power usage during video playback). We haven't measured with the merged implementation yet.

We should at least have something more concrete if we're going to accept the regression.

Compared the power consumption during video playback measured using an actual power meter. Got around 10% power savings during fullscreen video playback like the following.

  • PC1 intel GPU
    • Firefox Nightly with layer compositor: 29 - 30 W
    • Firefox Nightly with native compositor: 32 - 35 W
    • Chrome: 23 - 25 W
  • PC2 intel GPU
    • Firefox Nightly with layer compositor: 15 - 16 W
    • Firefox Nightly with native compositor: 17 - 18 W
    • Chrome: 15 - 16 W
  • PC3 AMD GPU
    • Firefox Nightly with layer compositor: 11 - 12 W
    • Firefox Nightly with native compositor: 12 - 13 W
    • Chrome: 11 - 12 W

Then with layer compositor, we could get the following advantages.

:jrmuizel, can comment 10 be enough benefit of layer compositor?

Flags: needinfo?(jmuizelaar)
Depends on: 2008399

(In reply to Sotaro Ikeda [:sotaro] from comment #11)

:jrmuizel, can comment 10 be enough benefit of layer compositor?

:jrmuizel answered like the following.


yeah, that's somewhat compelling

Flags: needinfo?(jmuizelaar)
Blocks: 1991874
Pushed by sikeda.birchill@mozilla.com: https://github.com/mozilla-firefox/firefox/commit/4294a1440f30 https://hg.mozilla.org/integration/autoland/rev/b184b5ea2fe4 Let WebRender layer compositor ride the trains to release on Windows r=gfx-reviewers,lsalzman
Status: REOPENED → RESOLVED
Closed: 2 months ago7 days ago
Resolution: --- → FIXED
Target Milestone: --- → 149 Branch

Is that something that should be mentioned in our 149 release notes? Thanks.

Flags: needinfo?(sotaro.ikeda.g)

(In reply to Pascal Chevrel:pascalc from comment #15)

Is that something that should be mentioned in our 149 release notes? Thanks.

Yes.

Flags: needinfo?(sotaro.ikeda.g)

Release Note Request (optional, but appreciated)
[Why is this notable]: WebRender rendering is changed from native compositor to layer compositor. It could reduce power usage of fullscreen video playback. And it could improve WebGL/WebGPU performance.
[Affects Firefox for Android]: No
[Suggested wording]: WebRender layer compositor is enabled on Windows. It may affect to WebRender performance.
[Links (documentation, blog post, etc)]:

relnote-firefox: --- → ?

Note added with this wording:

The WebRender layer compositor is now enabled on Windows, reducing power usage during full-screen video playback and improving WebGL/WebGPU performance.

I am keeping the relnote-firefox? flag open until we ship and include that in the final notes. Thanks

You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: