Bug 261892 – [GStreamer][MSE] video playback uses GstVA, except on YouTube

• Component: Media
• Status: RESOLVED FIXED

[Johannes Penßel]

Created attachment 467816 [details]
"GST_DEBUG=va*:6 epiphany": log of YT playback

WebKitGTK 2.42.0 happily plays VAAPI-accelerated video on any website in any HW-supported format, except on YouTube (tested with vp9 and av1). Tested with GNOME Web and Surf, on GNOME 45 in wayland and X11 sessions. If I remove the libraries for the GStreamer av1/vpx software decoders, playback on YT stops working altogether.

For reference, this testcase: https://tools.woolyss.com/html5-audio-video-tester/?u=woolyss.com/f/spring-vp9-vorbis.webm works flawlessly. (verified via intel_gpu_top)

Epiphany 44.0 (installed via flatpak) is not affected, but Epiphany.Devel 45.0 is. Firefox 117.0 works.

environment:
GST_PLUGIN_FEATURE_RANK=vaav1dec:MAX,vah264dec:MAX,vah265dec:MAX,vampeg2dec:MAX,vavp8dec:MAX,vavp9dec:MAX,vacompositor:MAX,vadeinterlace:MAX,vajpegdec:MAX,vapostproc:MAX,vah264lpenc:MAX,vah265lpenc:MAX

other software:
GStreamer 1.22.6
Mesa 23.2-rc3
libva-1.19
intel-media-driver-23.3.4

[Philippe Normand]

GST_DEBUG=3,va*:6,webkit*:6 log please :)

[Johannes Penßel]

Created attachment 467848 [details]
playback of vp9 content on YT with GST_DEBUG=3,va*:6,webkit*:6

With GST_DEBUG=3,va*:6,webkit*:6, the following errors are being logged:

0:00:02.157503867     2 0x55adb427d610 ERROR                default gstpipewirecore.c:91:make_core: (null)
0:00:02.158037939     2 0x55adb427d610 ERROR               pipewire gstpipewiredeviceprovider.c:625:gst_pipewire_device_provider_start:<pipewiredeviceprovider0> Failed to connect
0:00:02.307123798     2 0x55adb427d610 ERROR                GST_BUS gstbus.c:945:gst_bus_add_watch_full_unlocked:<bus95> Tried to add new watch while one was already there

I'm on pipewire 0.3.80 and alsa 1.2.10.

[Johannes Penßel]

Created attachment 467849 [details]
playback of vp9 content from a different source GST_DEBUG=3,va*:6,webkit*:6

For reference, this is a log captured while playing vp9 video from a different website. (with working HW accel)

[Philippe Normand]

Are the pipewire errors always and consistently happening when attempting to play YT videos?

[Johannes Penßel]

Yes, they're happening 100% predictably. Same goes for the flatpak version of GNOME Web 45.

[Philippe Normand]

I guess you don't have PulseAudio and you're using PipeWire directly?

If so, can you add this to GST_PLUGIN_FEATURE_RANK? pipewiresink:0

[Johannes Penßel]

Created attachment 467893 [details]
log with GST_PLUGIN_FEATURE_RANK+=pipewiresink:0

I use pipewire-pulse instead of pulseaudio-daemon, but libpulse is still present on my system. Is that what you mean? Sorry for the confusion. 

Seems like the same pipewire errors are still popping up with GST_PLUGIN_FEATURE_RANK+=pipewiresink:0, unfortunately.

[Philippe Normand]

and with pulsesink:0 ? Should fallback to alsasink if you have that.

[Johannes Penßel]

Created attachment 467899 [details]
log with GST_PLUGIN_FEATURE_RANK+=pulsesink:0

After replacing pipewire's soundserver with pulseaudio-daemon and disabling pipewire gstreamer support, I've finally managed to hit a different error message:

0:00:02.445350936     2 0x55e9c218f610 ERROR                   alsa gstalsadeviceprovider.c:57:add_device:<alsadeviceprovider0> Could not open device hw:0,0 for inspection!

No visible change in behavior otherwise though.

hw:0,0 is just my normal soundcard, which I've had zero issues with so far:

card 0: PCH [HDA Intel PCH], device 0: ALC256 Analog [ALC256 Analog]
  Subdevices: 1/1
  Subdevice #0: subdevice #0

[Philippe Normand]

I can reproduce this issue here. I think it's a GStreamer bug... For YT vp9 streams, decodebin3 doesn't parse the stream and attempts feed that to a compatible decoder, but vavp9dec requires a parsed stream, so it's not selected.

[Philippe Normand]

Can you try this workaround, downrank every vp9 decoder excepted the va one...

GST_PLUGIN_FEATURE_RANK=vp9dec:0,avdec_vp9:0

[Philippe Normand]

(In reply to Philippe Normand from comment #11)
> Can you try this workaround, downrank every vp9 decoder excepted the va
> one...
> 
> GST_PLUGIN_FEATURE_RANK=vp9dec:0,avdec_vp9:0

That should force decodebin3 to add a parsebin and then vavp9dec should be plugged...

[Philippe Normand]

https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/3024

[Johannes Penßel]

(In reply to Philippe Normand from comment #12)
> (In reply to Philippe Normand from comment #11)
> > Can you try this workaround, downrank every vp9 decoder excepted the va
> > one...
> > 
> > GST_PLUGIN_FEATURE_RANK=vp9dec:0,avdec_vp9:0
> 
> That should force decodebin3 to add a parsebin and then vavp9dec should be
> plugged...

That works! Dammit, I thought I had discovered an actual WebKit bug for once... Anyways, thank you very much for your help!

[Philippe Normand]

We shouldn't disable urisourcebin:parse-streams for MSE...

[Philippe Normand]

Pull request: https://github.com/WebKit/WebKit/pull/18825

[EWS]

Committed 269068@main (d14f07282a9d): <https://commits.webkit.org/269068@main>

Reviewed commits have been landed. Closing PR #18825 and removing active labels.

[Radar WebKit Bug Importer]

<rdar://problem/116668157>