1572691 - 26.61 - 27.76% ts_paint / ts_paint_webext (linux64-shippable-qr) regression on push 38b37dfa1f333de0785a1282f487549992698477 (Tue August 6 2019)

Status: RESOLVED FIXED

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

Description

[Marian Raiciof [:marauder]]

Talos has detected a Firefox performance regression from push:

https://hg.mozilla.org/integration/autoland/pushloghtml?changeset=38b37dfa1f333de0785a1282f487549992698477

As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

28% ts_paint_webext linux64-shippable-qr opt e10s stylo 271.08 -> 346.33
27% ts_paint_webext linux64-shippable-qr opt e10s stylo 272.00 -> 346.00
27% ts_paint linux64-shippable-qr opt e10s stylo 271.17 -> 343.33
19% tp5n nonmain_startup_fileio windows10-64-shippable opt e10s stylo 1,450,150.58 -> 1,731,475.58

You can find links to graphs and comparison views for each of the above tests at: https://treeherder.mozilla.org/perf.html#/alerts?id=22370

On the page above you can see an alert for each affected platform as well as a link to a graph showing the history of scores for this test. There is also a link to a treeherder page showing the Talos jobs in a pushlog format.

To learn more about the regressing test(s), please see: https://wiki.mozilla.org/TestEngineering/Performance/Talos

For information on reproducing and debugging the regression, either on try or locally, see: https://wiki.mozilla.org/TestEngineering/Performance/Talos/Running

*** Please let us know your plans within 3 business days, or the offending patch(es) will be backed out! ***

Our wiki page outlines the common responses and expectations: https://wiki.mozilla.org/TestEngineering/Performance/Talos/RegressionBugsHandling

Comment 1

[Martin Stránský [:stransky] (ni? me)]

Can you try to revert that patch? I wonder how this can affect the:

19% tp5n nonmain_startup_fileio windows10-64-shippable opt e10s stylo 1,450,150.58 -> 1,731,475.58

as this code is linux only and it's not even compiled for windows.

Comment 2

[Karl Tomlinson (:karlt)]

I triggered https://treeherder.mozilla.org/#/jobs?repo=autoland&revision=7168c4343dfa94e897d110b25cd62e5b655568e5&selectedJob=261062023
which makes it clear that "tp5n nonmain_startup_fileio windows10-64-shippable opt e10s stylo" is not affected by 38b37dfa1f333de0785a1282f487549992698477

Marian, are you able to unassign that regression from https://treeherder.mozilla.org/perf.html#/alerts?id=22370 and ensure a separate bug exists, please?

The graph supports the conclusion that the ts_paint linux64-shippable-qr regression is triggered by 38b37dfa1f333de0785a1282f487549992698477.

Comment 3

[Marian Raiciof [:marauder]]

Hi Karl, Martin,

I have unassign that alert and will open a separate bug for it.

Thank you!

Comment 4

[Karl Tomlinson (:karlt)]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=2d94750feacaaec2dc17cf9ac695f65c1516e32a and https://treeherder.mozilla.org/perf.html#/compare?originalProject=autoland&originalRevision=38b37dfa1f333de0785a1282f487549992698477&newProject=try&newRevision=2d94750feacaaec2dc17cf9ac695f65c1516e32a indicate that initializing mIsTransparent to true in Create() works around the webrender regression.

That would have the opposite effect when not drawing in the titlebar.

There's no difference in behavior without webrender, so I don't yet know why webrender is special.

(Bug 1572116 also indicates something unexpected in drawing behavior, and so could be related, but that seems to be only when webrender is off.)

Comment 5

[Karl Tomlinson (:karlt)]

https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=f866c1e6660a2cab858f93685cb5cfc39d32cb81&newProject=try&newRevision=6852198cf76abc21a113ab0f58178e556fccc950&framework=1
compares performance with PresShell::ComputeBackstopColor() overridden to return either always white or always transparent.

The comparison indicates that webrender performance is significantly better with transparent backstop than with white backstop and this difference explains the regression reported in this bug. Non-webrender performance does not show the same difference.

38b37dfa1f333de0785a1282f487549992698477 has been backed out, for other reasons, and so the performance regression will appear to be corrected.

Transparent backstop, however, is an accident waiting to be fixed. (Talos does not draw tabs in the titlebar, possibly based on GTK version.) It is likely to be corrected again.

Comment 6

[Timothy Nikkel (:tnikkel)]

(In reply to Karl Tomlinson (:karlt) from comment #5)

The comparison indicates that webrender performance is significantly better with transparent backstop than with white backstop and this difference explains the regression reported in this bug. Non-webrender performance does not show the same difference.

Transparent backstop, however, is an accident waiting to be fixed. (Talos does not draw tabs in the background, possibly based on GTK version.) It is likely to be corrected again.

Do you want to file a bug for this? A new bug is much more likely to get addressed then this bug which will likely be closed as fixed since the backout.

Comment 7

[Marian Raiciof [:marauder]]

Thank you Karl!
I'll mark this ticket as fixed since the 38b37dfa1f333de0785a1282f487549992698477 changeset has been backed out.