1908359 - 3.35 - 2.59% speedometer3 TodoMVC-Vue/DeletingAllItems/Sync / speedometer3 TodoMVC-Angular-Complex-DOM/CompletingAllItems/Sync (OSX) regression on Wed July 3 2024

Status: RESOLVED WONTFIX

(Core :: DOM: CSS Object Model, defect)

Description

[Florin Bilt]

Perfherder has detected a browsertime performance regression from push 782c9e2be6b8ee885a8d3a0294b60991f13cc12d. As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)	Performance Profiles
3%	speedometer3 TodoMVC-Vue/DeletingAllItems/Sync	macosx1015-64-nightlyasrelease-qr	fission webrender	5.55 -> 5.74	Before/After
3%	speedometer3 TodoMVC-Angular-Complex-DOM/CompletingAllItems/Sync	macosx1015-64-nightlyasrelease-qr	fission webrender	14.27 -> 14.64	Before/After

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests. Please follow our guide to handling regression bugs and let us know your plans within 3 business days, or the patch(es) may be backed out in accordance with our regression policy.

If you need the profiling jobs you can trigger them yourself from treeherder job view or ask a sheriff to do that for you.

You can run these tests on try with ./mach try perf --alert 1242

For more information on performance sheriffing please see our FAQ.

Comment 1

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

Set release status flags based on info from the regressing bug 1890842

Comment 2

[Daniel Holbert [:dholbert]]

I think Emily's semi-away for a few more days. In the meantime: emilio, maybe you could take a look here and see if you have any theories about how the patch could have regressed sp3 and what we might be able to do about it?

Comment 3

[Daniel Holbert [:dholbert]]

Looking more closely at comment 0 here, it looks like this specifically regressed speedometer3 for the "nightlyasrelease" build (at least, that's what's highlighted in comment 0 here). If this is really limited to that configuration, that's kinda interesting (and relevant to the eventual Firefox 130 release.)

(EDIT: it looks like maybe that was the only sp3 task that was run for this revision, so that might be the trivial explanation for why this was found on that config in particular. I've triggered some more sp3 tasks for this revision and its parent.)

Comment 4

[Daniel Holbert [:dholbert]]

Here's the compare view of this revision vs. its parent, filtered for "speedometer3":
https://treeherder.mozilla.org/perfherder/compare?originalProject=autoland&originalRevision=d65ed2688adc762c7c3ea34d5f397727410999db&newProject=autoland&newRevision=782c9e2be6b8ee885a8d3a0294b60991f13cc12d&framework=13&page=1&filter=speedometer3

You can click into "subtests" to see the per-subtest breakdown.

I spammed some retriggers to hopefully get better confidence / signal-to-noise here.

Comment 5

[Daniel Holbert [:dholbert]]

So far it looks like the retriggers are confirming the (small) regression in these subtests. As of right now with 25 runs before & 26 runs after:

• TodoMVC-Vue/DeletingAllItems/Sync: 3.75% regression
  https://treeherder.mozilla.org/perfherder/comparesubtest?originalProject=autoland&newProject=autoland&newRevision=782c9e2be6b8ee885a8d3a0294b60991f13cc12d&originalSignature=4721628&newSignature=4721628&framework=13&application=firefox&originalRevision=d65ed2688adc762c7c3ea34d5f397727410999db&page=1&filter=TodoMVC-Vue%2FDeletingAllItems%2FSync

• TodoMVC-Angular-Complex-DOM/CompletingAllItems/Sync: 1.66% regression
  https://treeherder.mozilla.org/perfherder/comparesubtest?originalProject=autoland&newProject=autoland&newRevision=782c9e2be6b8ee885a8d3a0294b60991f13cc12d&originalSignature=4721628&newSignature=4721628&framework=13&application=firefox&originalRevision=d65ed2688adc762c7c3ea34d5f397727410999db&page=1&filter=TodoMVC-Angular-Complex-DOM%2FCompletingAllItems%2FSync

However, the subtests results show improvements in many other subtests, so I'm not sure to-what-extent these regressions (and the other supposed improvements) are real vs. just noise.

Comment 6

[Emilio Cobos Álvarez (:emilio)]

My somewhat wild guess is that the compiler managed to devirtualize WrapObject and now no longer can, but that seems really far-fetched... I suspect it's just noise in practice... Florin, could we get some profiling runs by any chance?

Comment 7

[Emilio Cobos Álvarez (:emilio)]

Ah, daniel pointed out that there are profiles in comment 0!

Comment 8

[Emilio Cobos Álvarez (:emilio)]

I don't see any of the relevant code from the regressing bug getting called in those profiles...

Comment 9

[Emilio Cobos Álvarez (:emilio)]

Closest thing is set CSS2Properties.display...

Comment 10

[Daniel Holbert [:dholbert]]

On this macosx1015-64-nightlyasrelease-qr platform, the speedometer3 score overall got very-slightly-better with Emily's change (as of current results shown in perfherder compare view, with 45 runs for "before" & 46 runs "after") -- there's a 0.07% increase (improvement) in the speedometer3 score (a negligible-but-positive overall improvement):
https://treeherder.mozilla.org/perfherder/compare?originalProject=autoland&originalRevision=d65ed2688adc762c7c3ea34d5f397727410999db&newProject=autoland&newRevision=782c9e2be6b8ee885a8d3a0294b60991f13cc12d&framework=13&page=1&filter=macosx1015-64-nightlyasrelease-qr+speedometer3

This negligible improvement breaks down into small wins and small losses on various subtests, which might be due to PGO differences, with no obvious identifiable reason for any relevant-to-speedometer behavior to have changed, or for the regression in these two subtest-in-particular, per comment 8.

Given the above, emilio and I agree that this isn't likely to be a valuable thing to spend time investigating much further.