2.93% raptor-tp6-reddit-firefox-cold fcp (linux64-shippable-qr) regression on push 6fb0f25b2f5d3fd39a53d304674e91f24ffaa259 (Sat May 16 2020)
Categories
(Core :: Graphics: WebRender, defect)
Tracking
()
Tracking | Status | |
---|---|---|
firefox76 | --- | unaffected |
firefox77 | --- | unaffected |
firefox78 | --- | wontfix |
People
(Reporter: Bebe, Unassigned)
References
(Regression)
Details
(Keywords: perf, perf-alert, regression)
Perfherder has detected a raptor performance regression from push 6fb0f25b2f5d3fd39a53d304674e91f24ffaa259. As author of one of the patches included in that push, we need your help to address this regression.
Regressions:
3% raptor-tp6-reddit-firefox-cold fcp linux64-shippable-qr opt 275.67 -> 283.75
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 offending patch(es) will be backed out in accordance with our regression policy.
For more information on performance sheriffing please see our FAQ.
Reporter | ||
Updated•4 years ago
|
Comment 1•4 years ago
|
||
Set release status flags based on info from the regressing bug 1635610
Comment 2•4 years ago
|
||
The push https://hg.mozilla.org/integration/autoland/pushloghtml?changeset=6fb0f25b2f5d3fd39a53d304674e91f24ffaa259 doesn't appear to be relevant, is this a mistake?
Comment 3•4 years ago
|
||
As Dzmitry mentioned above, that push range doesn't seem to have any commits related to WR code changes, but perhaps it's just the wrong link?
I'm on PTO today, but if there is an updated link with a push from me, I'll take a look at it tomorrow (please NI me again if that's the case).
Reporter | ||
Comment 4•4 years ago
|
||
Sorry for the confusion I reassigned the alert to the right commit:
== Change summary for alert #26048 (as of Tue, 26 May 2020 07:30:09 GMT) ==
Regressions:
3% raptor-tp6-reddit-firefox-cold fcp linux64-shippable-qr opt 275.67 -> 283.75
Improvements:
1% raptor-tp6-reddit-firefox-cold fcp linux64-shippable-qr opt 287.00 -> 283.42
For up to date results, see: https://treeherder.mozilla.org/perf.html#/alerts?id=26048
Comment 5•4 years ago
|
||
I compared which shaders get compiled before and after this patch - there is one extra shader being compiled, with signature:
{ base_filename: "ps_text_run", features: ["GLYPH_TRANSFORM", "ALPHA_PASS"], source_type: Optimized(Gl) }
So with this change, we end up spending an extra 6-8ms on cold start compiling a shader that wasn't previously required.
We don't have shader caching on Linux (yet), which is why we see this show up here. On Windows (and Android) we cache shaders after the first use, which minimizes this problem.
So, this is expected, for now, at least and I think we can close it. I am currently working on some text run optimization work in [1] which may reduce the number of text shader variants we need, although that's not strictly related to this.
Updated•4 years ago
|
Updated•4 years ago
|
Comment 6•2 years ago
|
||
Too old to fix now.
Description
•