Gnome Wayland, Debian Testing, Intel Iris Graphics 6100 (BDW GT3), intel-media-va-driver-non-free 22.1.0+ds1-1

MOZ_DISABLE_RDD_SANDBOX=1 mozregression --good 2022-01-20 --bad 2022-01-28 --pref media.ffmpeg.vaapi.enabled:true -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -P stdout

8:50.27 INFO: Last good revision: 5834c469b533e08bbd08a8ffbaad1388e5d61344
8:50.27 INFO: First bad revision: c3d14f2b234a7494edf3231113af52f5fc2d2355
8:50.27 INFO: Pushlog:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=5834c469b533e08bbd08a8ffbaad1388e5d61344&tochange=c3d14f2b234a7494edf3231113af52f5fc2d2355

c3d14f2b234a7494edf3231113af52f5fc2d2355 stransky — Bug 1724385 [Linux] Try to create EGLImage over decoded VA-API video frame r=alwu,emilio,media-playback-reviewers

VAAPI is used in both cases according to intel_gpu_top:

• dmabuf textures enabled = glitches
  MOZ_DISABLE_RDD_SANDBOX=1 mozregression --launch 2022-01-28 --pref media.ffmpeg.vaapi.enabled:true media.ffmpeg.dmabuf-textures.disabled:false -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -P stdout

• dmabuf textures disabled = no glitches
  MOZ_DISABLE_RDD_SANDBOX=1 mozregression --launch 2022-01-28 --pref media.ffmpeg.vaapi.enabled:true media.ffmpeg.dmabuf-textures.disabled:true -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -P stdout

This could be missing fencing?

Now I uninstalled intel-media-va-driver-non-free and the glitch does not occur when actually using SW decoding to dmabuf texture:

MOZ_DISABLE_RDD_SANDBOX=1 MOZ_LOG="Dmabuf:5, PlatformDecoderModule:5" mozregression --launch 2022-01-28 --pref media.ffmpeg.vaapi.enabled:true media.ffmpeg.dmabuf-textures.disabled:false -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -P stdout

0:34.33 INFO: b'[RDD 50711: MediaPDecoder #1]: D/PlatformDecoderModule FFMPEG: DMABuf Got one frame output with pts=3680000dts=3720000 duration=40000 opaque=-9223372036854775808'

MOZ_DISABLE_RDD_SANDBOX=1 MOZ_LOG="Dmabuf:5, PlatformDecoderModule:5" mozregression --launch 2022-01-28 --pref media.ffmpeg.vaapi.enabled:true -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -P stdout > 1752605_vaapi_glitches.txt

I think it may come from https://phabricator.services.mozilla.com/D136726
I'll create a test try build next week with that reverted.

(In reply to Martin Stránský [:stransky] (ni? me) from comment #5)

I think it may come from https://phabricator.services.mozilla.com/D136726
I'll create a test try build next week with that reverted.

That patch has its own and slightly earlier autoland push (bug 1724385 comment 26) which seems to be unaffected:
MOZ_DISABLE_RDD_SANDBOX=1 mozregression --repo autoland --launch 32af039c697d --pref media.ffmpeg.vaapi.enabled:true -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -a https://upload.wikimedia.org/wikipedia/commons/1/18/The_Earth_in_4k.webm -P stdout

0:15.56 INFO: Running autoland build built on 2022-01-25 08:31:52.413000, revision 32af039c

Latest m-c is still affected:
MOZ_DISABLE_RDD_SANDBOX=1 mozregression --launch 48e8fb0b62c587e64b43e9d51a0dff7fdf4b7b2c --pref media.ffmpeg.vaapi.enabled:true -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -a https://upload.wikimedia.org/wikipedia/commons/1/18/The_Earth_in_4k.webm -P stdout

This bug seems to be fixed (and VAAPI still works) in try builds from

• bug 1752097 comment 6:
  MOZ_DISABLE_RDD_SANDBOX=1 mozregression --repo try --launch 400cffc73157869a0eb47d45e86377e9397ecb89 --pref media.ffmpeg.vaapi.enabled:true -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -a https://upload.wikimedia.org/wikipedia/commons/1/18/The_Earth_in_4k.webm -P stdout
• bug 1752282 comment 17:
  MOZ_DISABLE_RDD_SANDBOX=1 mozregression --repo try --launch f7506cb0966ca8e20c14f8eff1db74b33a83520e --pref media.ffmpeg.vaapi.enabled:true -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -a https://upload.wikimedia.org/wikipedia/commons/1/18/The_Earth_in_4k.webm -P stdout

Thanks!