Closed Bug 1748093 Opened 2 years ago Closed 2 years ago

video playback slow and choppy since firefox 0.93

Categories

(Core :: Graphics: WebRender, defect)

Firefox 92
defect

Tracking

()

RESOLVED WORKSFORME

People

(Reporter: eric.brunet, Unassigned)

References

(Regression)

Details

(Keywords: regression)

User Agent: Mozilla/5.0 (X11; Fedora; Linux x86_64; rv:92.0) Gecko/20100101 Firefox/92.0

Steps to reproduce:

I am using fedora 34 linux on a recent Dell laptop (XPS 13), with Xorg and kde (whereas the default is wayland and gnome). The screen has a resolution 3840x2160.
Playing video from youtube at full HD used to pose no problem, but it has become choppy since firefox 0.93 (by which I mean the program distributed by fedora in the package firefox-93.0-2.fc34.x86_64.rpm). The problem is present with firefox 0.93, 0.94 and 0.95. I am stuck with firefox 0.92 (firefox-92.0-2.fc34.x86_64.rpm) in order to have correct (and, actually, perfect) video playback.

The problem occurs with a fresh profile. I have tried two drivers for Xorg: the intel one, and the modeset one. The problem occurs nearly all the time, but I have seen it working nicely occasionnaly in versions 0.93 to 0.95.

When the problem occurs, here is what "top" displays:

172898 eric 20 0 4577856 961728 445800 S 201,3 6,0 1:44.72 GeckoMain
173483 eric 20 0 3127296 339452 115424 S 32,6 2,1 0:29.59 Isolated Web Co
173577 eric 20 0 303724 80512 59856 S 23,9 0,5 0:10.16 RDD Process
170865 root 20 0 1386216 311280 279028 S 7,3 1,9 0:11.96 Xorg

(Notice GeckoMain at 200% cpu). When the problem does not occur while playing a video, here is what top displays:

175294 eric 20 0 256420 76536 56416 S 31,8 0,5 0:22.34 RDD Process
175190 eric 20 0 3086980 307036 115260 S 30,5 1,9 0:54.05 Isolated Web Co
174638 eric 20 0 4561284 516784 242720 S 22,8 3,2 1:04.50 firefox
171235 eric 20 0 2200892 140284 98932 S 12,6 0,9 0:30.12 kwin_x11

When starting firefox ≥ 0.93, I see the message:

ATTENTION: default value of option mesa_glthread overridden by environment.

I don't know if this is relevant.

Nota: I'd like to give more information, I am sure there are some built-in tracing and logging capablity, but I don't know how to use it. What test could I do to help?

Actual results:

Video playback is slow for recent versions of firefox

Expected results:

Video playback should play at full speed.

The Bugbug bot thinks this bug should belong to the 'Core::Widget: Gtk' component, and is moving the bug to that component. Please revert this change in case you think the bot is wrong.

Component: Untriaged → Widget: Gtk
Product: Firefox → Core

The issue is still present with firefox as present in the package firefox-96.0.3-1.fc34
I have just tried it today, and promptly reverted back to firefox 0.92 after trying a youtube video.

Since this is reportedly a regression, can you run the following to figure out what broke this for you?

Though maybe Fedora ships with vaapi support enabled in which case even the 92 version could be choppy (in that case, there are other commands that would allow to figure out what's the culprit). But anyways let's try this for now:

$ pip3 install --user mozregression
$ mozregression --good 92 -a https://youtube.com # Url here could be a specific youtube video for one less click while bisecting

That should end up with a revision range that broke this.

Flags: needinfo?(eric.brunet)

(Thanks in advance, btw)

Thank you for your answer. The mozregression tool is impressive!
I could make a bissection:

48:09.09 INFO: Last good revision: 5938c983ab44a2b494e0a3e41f11dc2541016a37
48:09.09 INFO: First bad revision: 6c779cb76a8f5f68c6f7b95d12b311fa305a4f71
48:09.09 INFO: Pushlog:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=5938c983ab44a2b494e0a3e41f11dc2541016a37&tochange=6c779cb76a8f5f68c6f7b95d12b311fa305a4f71

which leads to the commit:
Robert Mader — Bug 1710400 - Blocklist HW-WR on Intel DDX driver, r=aosmond

Ok, so this makes sense: I am primarly using the intel driver, hw acceleration has been disabled, and my video is choppy.

I went back to the modeset driver, and I can no longer replicate the issue. (Which is strange, because I think I did try that before submitting this bug report and saw the issue also with modeset.)

I am a bit annoyed because the intel driver works better for me then the modeset driver in some other aspects. I imagine there's no way to get back hw acceleration with the intel driver?

Well, maybe it is time I give a serious try to wayland...

Alternatively you can opt into hardware acceleration with gfx.webrender.all=true in about:config , if you're not seeing the issues that that bug was supposed to fix.

Component: Widget: Gtk → Graphics: WebRender
Regressed by: 1710400

Yes, thank you, everything works perfectly now, even with the intel driver.

Status: UNCONFIRMED → RESOLVED
Closed: 2 years ago
Flags: needinfo?(eric.brunet)
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.