17.77 - 17.53% ts_paint + 1 more (Linux) regression on Wed April 29 2026
Categories
(Core :: Audio/Video: Playback, defect)
Tracking
()
| Tracking | Status | |
|---|---|---|
| firefox-esr115 | --- | unaffected |
| firefox-esr140 | --- | unaffected |
| firefox150 | --- | unaffected |
| firefox151 | --- | unaffected |
| firefox152 | --- | fixed |
People
(Reporter: intermittent-bug-filer, Assigned: tboiko)
References
(Regression)
Details
(4 keywords)
Attachments
(2 files)
Perfherder has detected a talos performance regression from push f6affd94efbcb7de652df67b08e4ad52c437743b. 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] |
|---|---|---|---|---|
| 18% | ts_paint (doc) | linux2404-64-shippable | e10s fission stylo webrender-sw | 423.08 -> 498.25 |
| 18% | ts_paint_webext (doc) | linux2404-64-shippable | e10s fission stylo webrender-sw | 422.58 -> 496.67 |
Support & Actionable Steps
For questions, please contact afinder@mozilla.com. You can also find help on Slack (#perf-help) or Matrix (#perftest).
Action guide details:
- See the guide to handling regression bugs for investigation steps.
- Review the alert summary for affected tests, graphs, and comparisons.
- Check the PerfCompare results.
- Tests on Try: Run all tests using
./mach try perf --alert 49998(See documentation link for details). - Profiling jobs: Trigger jobs from Treeherder job view or request them from afinder@mozilla.com.
| Assignee | ||
Comment 1•6 days ago
|
||
At the moment I think that linux startup/paint regressed because Vulkan video capability probing runs an extra synchronous vulkantest subprocess during early gfx/video initialization, same class of work as existing vaapitest probing but additive on the critical path.
| Assignee | ||
Comment 2•6 days ago
|
||
However, media.hardware-video-decoding-vulkan.enabled is false by default, so the vulkantest should not be triggered.
(Must be true for FEATURE_HARDWARE_VIDEO_DECODING_VULKAN to call GetDataVulkan() (and thus run vulkantest)).
| Assignee | ||
Comment 3•5 days ago
|
||
DMABuf:
-
AppendDmaBufModifiersFromEGLIfEmpty() was called unconditionally from
SetModifiersToGfxVars() on X11 startup for P010 and NV12 formats,
triggering EGL initialisation and two driver round-trips per format
(~80 ms total) even when hardware video decoding is disabled. -
Move the queries into a new AppendEGLVideoModifiers() method and call it
from InitPlatformHardwareVideoConfig() only when featureDec.IsEnabled(),
so machines without hardware video decoding (the default) pay no cost.
Measured ts_paint regression: +75 ms on Linux (Bug 2036839).
Updated•5 days ago
|
| Assignee | ||
Comment 4•5 days ago
|
||
Most likely AppendDmaBufModifiersFromEGLIfEmpty() is called unconditionally from SetModifiersToGfxVars() on X11 startup for P010 and NV12, triggering EGL initialisation and two driver round-trips even when hardware video decoding is disabled. Provided a patch, please try if the above patch helps
Comment 7•5 days ago
|
||
Backed out for causing bustages in DMABufFormats.cpp.
Backout link: https://hg.mozilla.org/integration/autoland/rev/c275d15ee06b
| Assignee | ||
Comment 8•5 days ago
|
||
Let me take a look at this fix, I'll submit an updated version in the next few hours.
| Assignee | ||
Comment 9•5 days ago
|
||
DMABuf:
-
AppendDmaBufModifiersFromEGLIfEmpty() was called unconditionally from
SetModifiersToGfxVars() on X11 startup for P010 and NV12 formats,
triggering EGL initialisation and two driver round-trips per format
(~80 ms total) even when hardware video decoding is disabled. -
Move the queries into a new AppendEGLVideoModifiers() method and call it
from InitPlatformHardwareVideoConfig() only when featureDec.IsEnabled(),
so machines without hardware video decoding (the default) pay no cost.
Measured ts_paint regression: +75 ms on Linux (Bug 2036839).
Comment 10•5 days ago
|
||
Great. Tymur, is it the final version I shall review and land or are you going to do more changes here?
Thanks.
| Assignee | ||
Comment 12•5 days ago
|
||
Sorry, my last submitted version created a duplicate revision.
The latest version to land is D298517.
Comment 13•5 days ago
|
||
Comment 14•5 days ago
|
||
| bugherder | ||
Updated•4 days ago
|
Comment 15•4 days ago
|
||
Authored by https://github.com/t-boiko
https://github.com/mozilla/enterprise-firefox/commit/04a81dc2cdeb229586c2826295d3bf4a6b2858d7
[enterprise-main] Bug 2036839 - DMABuf - Defer EGL modifier queries to after hardware video decoding is enabled. r=stransky
Authored by Atila Butkovits
https://github.com/mozilla/enterprise-firefox/commit/5e7ae585d5a964f34c38dac2393c8ae488d6e1e6
[enterprise-main] Revert "Bug 2036839 - DMABuf - Defer EGL modifier queries to after hardware video decoding is enabled. r=stransky" for causing bustages in DMABufFormats.cpp.
Authored by https://github.com/t-boiko
https://github.com/mozilla/enterprise-firefox/commit/92ab17043ca3a8c384856180acc8b722cceb99dc
[enterprise-main] Bug 2036839 - DMABuf - Defer EGL modifier queries to after hardware video decoding is enabled. r=stransky
Description
•