1701395 - performance regression in chalkboard benchmark (https://testdrive-archive.azurewebsites.net/performance/chalkboard/)

Status: RESOLVED FIXED

(Core :: Graphics: Canvas2D, defect, P1)

Description

[furkan]

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:89.0) Gecko/20100101 Firefox/89.0

Steps to reproduce:

1. go to https://testdrive-archive.azurewebsites.net/performance/chalkboard/ run a benchmark and compare the results with a nightly before https://bugzilla.mozilla.org/show_bug.cgi?id=1699790

Actual results:

https://bugzilla.mozilla.org/show_bug.cgi?id=1699790 has landed. I did a mozregression and found this https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=d08490cb7ac70c79a200ad93effef91d2570f793&tochange=8350638712cf491cbdabc5ed1cf2f72139daed12

Expected results:

https://www.vsynctester.com/gputest.html benchmark should have improved without degrading performance of chalkboard test. I also tested the vsynctester with and without https://bugzilla.mozilla.org/show_bug.cgi?id=1699790 I got similar results where in chalkboard I get worse results.

Comment 1

[furkan]

On my machine the results got from ~7.50 seconds to ~11.80. I also requested in the Nightly Matrix chat from users to see how their results are:

smowtenshi: reported from 6.41 to 8.39
B.J. Herbison: reported from 6.72 to 11.52

Comment 2

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

Set release status flags based on info from the regressing bug 1699790

Comment 3

[Bob Owen (:bobowen)]

Attachment: Bug 1701395: Use GetMaxTextureSizeForFeatureLevel for D3D11 canvas texture size checks. r=jrmuizel!

This also sets the webrender max_texture_size back to 8192, to fix performance
regression.

Comment 4

[Bob Owen (:bobowen)]

What do you think would be the best approach here?
I guess I could try and add the feature level to TextureFactoryIdentifier, but I can't quite find my way round the rust code to work out how to populate it.

Comment 5

[Nicolas Silva [:nical]]

Attachment: Bug 1701395 - Report the device's max texture size (up to 16k). r=jrmuizel

This patch separates the max_texture_size reported by webrender's renderer (limit to what webrender can consume) from the max texture size used internally as a threshold for tiling.

Comment 6

[Bob Owen (:bobowen)]

https://treeherder.mozilla.org/jobs?repo=try&revision=7a82e56f21f44d55d09109a44a36ff20f4cd186d

Comment 7

[Bob Owen (:bobowen)]

Another try push with the wrench test fixed:
https://treeherder.mozilla.org/jobs?repo=try&revision=d4d21c1d479b4217158f99593dcc638bb1fc96a5

Comment 8

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/7ede736ae56e
Report the device's max texture size (up to 16k). r=jrmuizel

Comment 9

[Bogdan Tara[:bogdan_tara | bogdant]]

https://hg.mozilla.org/mozilla-central/rev/7ede736ae56e