1880018 - [gpu-canvas] Performance regression on a Codepen demo (https://codepen.io/KilledByAPixel/pen/WNQoMYg) with gpu-canvas

Status: VERIFIED FIXED

(Core :: Graphics: Canvas2D, defect)

Description

[Mayank Bansal]

Use latest nightly on Windows.

Set the following prefs
gfx.direct2d.disabled = true
gfx.canvas.accelerated = true
gfx.canvas.remote.worker-threads = 0

Go to https://codepen.io/KilledByAPixel/pen/WNQoMYg

AR: The browser freezes and the demo is very slow to run.

Regression range:
Bug 1829026 - Handle KERN_ABORTED from semaphore_wait. r=aosmond
Differential Revision: https://phabricator.services.mozilla.com/D195952

Note that the performance regressed for both GPU-canvas and d2d-canvas, but gpu-canvas has a much worse regression. With skia-canvas, it runs absolutely fine. I have not filed any bug for d2d-canvas. Let me know if needed.

Comment 1

[Mayank Bansal]

GPu-canvas: https://share.firefox.dev/49cxanz (very bad)
skia-canvas: https://share.firefox.dev/49pAbkg (very good)
d2d-canvas: https://share.firefox.dev/4bBO2Wd (bad)

Comment 2

[Mayank Bansal]

Attachment: about:support

Comment 3

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1829026

Comment 4

[Lee Salzman [:lsalzman]]

Attachment: Bug 1880018 - Implement RecordedExtractSubrect to forward ExtractSubract calls for SourceSurfaceRecording. r?aosmond

ExtractSubrect calls were not getting forwarded on SourceSurfaceRecording, causing us to do unnecessary readbacks
when hitting the CopySurface fallback path. SourceSurfaceWebgl actually implements ExtractSubrect already, but it
was never being used since the switch over to the DrawTargetRecording infrastructure.

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/0dce8e9d0ebb
Implement RecordedExtractSubrect to forward ExtractSubract calls for SourceSurfaceRecording. r=aosmond

Comment 6

[Serban Stanca [:SerbanS]]

https://hg.mozilla.org/mozilla-central/rev/0dce8e9d0ebb

Comment 7

[Mayank Bansal]

Profile from latest Nightly: https://share.firefox.dev/3uxavDo
Subjectively it doesnt feel much better.

Comment 8

[Alexandru Trif, Desktop Test Engineering [:atrif]]

Reproduced the issue with Firefox 124.0a1 (2024-02-13) on Windows 10x64. The animation is laggy and the browser freezes.
I can no longer reproduce the freeze with Firefox 124.0b2 on Windows 10x64, macOS 13 and Ubuntu 23.1 However, the animation is not as smooth as before bug 1829026 with the same steps.
Also, it looks like the browser shows a slight delay when playing the animation on Ubuntu 23.1 and macOS 13 ARM. Should we open another issue for the remaining problem?

Comment 9

[Lee Salzman [:lsalzman]]

For now I worked on the actionable things in this bug, and there are probably enough existing perf bugs to cover the remaining.

Comment 10

[Alexandru Trif, Desktop Test Engineering [:atrif]]

Thank you! Closing this as verified per comment 8 and comment 9 since the freeze is no longer reproducible.

Comment 11

[Mayank Bansal]

This has improved : https://share.firefox.dev/462PSgB
blaming bug 1907755