Closed Bug 2059433 Opened 3 months ago Closed 3 months ago

19.91% pdfpaint bug857031.pdf (OSX) regression on Thu July 23 2026

Categories

(Core :: Graphics: Canvas2D, defect, P3)

defect

Tracking

()

RESOLVED WONTFIX
Tracking Status
firefox-esr140 --- unaffected
firefox-esr153 --- unaffected
firefox153 --- unaffected
firefox154 --- unaffected
firefox155 --- fix-optional

People

(Reporter: performance-sheriff-bot, Assigned: jnicol)

References

(Regression)

Details

(4 keywords)

Perfherder has detected a talos performance regression from push c9c78ae99215d977e1b6c17377637ae04f81275f. As author of one of the patches included in that push, we need your help to address this regression.

Please acknowledge, and begin investigating this alert within 3 business days, or the patch(es) may be backed out in accordance with our regression policy.

Regression Test Platform Options Absolute values [old vs new]
20% pdfpaint bug857031.pdf (doc) macosx1470-64-shippable e10s fission stylo webrender 574.31 ms -> 688.64 ms
Improvement Test Platform Options Absolute values [old vs new]
21% pdfpaint issue18444.pdf (doc) macosx1470-64-shippable e10s fission stylo webrender 747.93 ms -> 589.64 ms
18% pdfpaint ichiji.pdf (doc) macosx1470-64-shippable e10s fission stylo webrender 650.47 ms -> 536.50 ms
14% pdfpaint issue15833.pdf (doc) macosx1470-64-shippable e10s fission stylo webrender 585.07 ms -> 505.54 ms
11% pdfpaint issue16384.pdf (doc) macosx1470-64-shippable e10s fission stylo webrender 367.14 ms -> 328.12 ms

Support & Actionable Steps

For questions, please contact fbilt@mozilla.com. You can also find help on Slack (#perf-help) or Matrix (#perftest).

Action guide details:

Flags: needinfo?(jnicol)

Set release status flags based on info from the regressing bug 2055839

Here's a profile of current main: https://share.firefox.dev/4vUctXV
And after reverting "Bug 2055839 - Use shared PBO for DrawTargetWebgl uploads.": https://share.firefox.dev/4wyqNqn

(The other commit in the push won't have had any effect, as ANGLE is disabled by default on macos)

We can see that the commit did improve texture upload performance as was intended. UploadSurface accounts for 138 samples with the reversion, but only 40 samples on main. We can also see that with the shared PBO, we see quite a lot of skia work on the content process main thread, indicating we have fallen back from accelerated canvas. whereas without the shared PBO, there's no canvas work on the main thread, indicating we have not.

Also I can see when running locally that in the no shared PBO case, the PDF contents don't appear to always get drawn during each test iteration.

So my theory is that bug 2055839 improved performance enough so that that we now render enough frames for accelerated canvas to get disabled, then skia canvas takes over. Previously we didn't render enough frames to fall back, and the test measurement ended early for some reason.

I'll dig a bit more to make sure.

But in any case even if this were a bona fide regression, the improvements outweight the regression, so please don't back out.

Flags: needinfo?(jnicol)
Assignee: nobody → jnicol
Severity: -- → S3
Priority: -- → P3

It has been over 3 days with no activity on this performance regression.

:jnicol, since you are the author of the regressor, bug 2055839, which triggered this performance alert, could you please provide a progress update?

If this regression is something that fixes a bug, changes the baseline of the regression metrics, or otherwise will not be fixed, please consider closing it as WONTFIX. See this documentation for more information on how to handle regressions.

For additional information/help, please needinfo the performance sheriff who filed this alert (they can be found in comment #0), or reach out in #perftest, or #perfsheriffs on Element.

For more information, please visit BugBot documentation.

Flags: needinfo?(jnicol)

The issue appears to be that without the patch, accelerated canvas is so slow that it doesn't render enough frames to decide it wants to fall back to accelerated canvas. and the test measurement doesn't reflect the fact that the canvas render thread is still busy. with accelerated canvas it's fast enough that it decides it wants to fall back to unaccelerated, and then the main thread is slower because it's doing the rendering itself, which makes the test measurement take longer.

and in any case, the wins more than outweigh the reported regression.

Status: NEW → RESOLVED
Closed: 3 months ago
Flags: needinfo?(jnicol)
Resolution: --- → WONTFIX
You need to log in before you can comment on or make changes to this bug.