Even if a video is paused, the GPU process will still use a lot of CPU
Categories
(Core :: Graphics: WebRender, defect)
Tracking
()
People
(Reporter: tgnff242, Unassigned)
References
(Regression)
Details
(Keywords: regression)
User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:77.0) Gecko/20100101 Firefox/77.0
Steps to reproduce:
- Start Nightly with a clean profile.
- Open a few windows and start playing a video in them. It can be from a site or a local file, it doesn't matter.
- Pause the videos in all windows.
Actual results:
The GPU process uses a lot of CPU. The more videos in separate windows, the more severe the issue becomes.
Expected results:
This is a regression, but, unfortunately, there are builds that are crashing at startup in my mozregression run, so I can't find the exact commit. This is as close I could get: https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=47426d145e246fa1924fbda83a8ecb0d25a6f606&tochange=8854e1368b2088e6b4abb9db1054ae45cb22e5bf
I captured the following profile with five paused local videos in separate windows: https://profiler.firefox.com/public/7e497e90607814259f663bfb3133f90a0d708a94/calltree/?globalTrackOrder=0-1-2-3-4-5-6-7&hiddenGlobalTracks=1-2-3-4-5-6&hiddenLocalTracksByPid=155568-1&localTrackOrderByPid=155534-1-2-0~155880-0~156043-0~155564-0~155834-0~155732-0~155681-0~155568-2-0-1~&thread=0&v=4
Comment 1•5 years ago
|
||
Bugbug thinks this bug should belong to this component, but please revert this change in case of error.
Can you confirm that you're not using VA-API?
I'm not using VA-API. I'm not using Wayland at all.
Comment 4•5 years ago
|
||
Because this bug's Severity has not been changed from the default since it was filed, and it's Priority is -- (non,) indicating it has has not been previously triaged, the bug's Severity is being updated to -- (default, untriaged.)
Could you capture a profile using the 'Media' preset in the profiler? What method are you using to check the CPU usage?
Updated•5 years ago
|
I can't reproduce the issue on Nightly anymore. I ran mozregression with --find-fix and found Bug 1636038 as a fix, which means Bug 1612440 caused the regression. The beta version is still affected when WebRender is enabled, though.
Here is the perf profile with media preset on an affected Nightly build: https://perfht.ml/3dHakap I captured it while having five paused local (file:///) videos in separate windows. The GPU process was using about 12% of a 4-core/8-thread laptop hasswell CPU. It's not too high for local videos, but for yt videos it easily reaches 30%.
Thanks! Marking as a dupe of the bug above, as it sounds like that resolves the issue.
Description
•