1658772 - Mesa/Nouveau: Distorted video when VAAPI / X11 enabled

Status: RESOLVED DUPLICATE of bug 1724385

(Core :: Audio/Video: Playback, defect, P5)

Description

[Ilgaz Öcal]

Attachment: youtubeglitch.png

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:81.0) Gecko/20100101 Firefox/81.0

Steps to reproduce:

Install latest nightly
Enable ffmpeg-vaapi and webrender
Launch Firefox with EGL enabled
Play some video on Youtube

Actual results:

Video is distorted (mostly green) and if you keep playing it, it will hang the OS.

Expected results:

Play video with hardware acceleration

Comment 1

[Ilgaz Öcal]

Attachment: firefox about:support info

Console output has lines like:
[Child 2859: MediaPDecoder #2]: D/PlatformDecoderModule DMABUF/VA-API Got one frame output with pts=3128122dts=3128122 duration=41711 opaque=-9223372036854775808

and graphics info
 >>> inxi -xG                                                             [22]
Graphics:
  Device-1: NVIDIA C79 [GeForce 9400] vendor: Apple driver: nouveau 
  v: kernel bus ID: 02:00.0 
  Display: x11 server: X.org 1.20.8 driver: nouveau unloaded: modesetting 
  resolution: <xdpyinfo missing> 
  OpenGL: renderer: NVAC v: 3.3 Mesa 20.1.3 direct render: Yes

Comment 2

[BugBot [:suhaib / :marco/ :calixte]]

Bugbug thinks this bug should belong to this component, but please revert this change in case of error.

Comment 3

[Darkspirit]

It can look somewhat similar if one has two GPUs, but uses the wrong one to decode (bug 1622132 comment 0), but in this case only one GPU is present.

https://en.wikipedia.org/wiki/Nvidia_PureVideo#Table_of_GPUs_containing_a_PureVideo_SIP_block
https://www.techpowerup.com/gpu-specs/nvidia-c79.g412

VP3

https://nouveau.freedesktop.org/wiki/VideoAcceleration/

H264 DONE
H.264 videos of movie trailers often end up rendering with some artifacts. Never seen in an actual movie though.

There is a first/open MR from Nvidia, don't know if related at all: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3724

Comment 4

[Darkspirit]

There is one crash report mentioning DMABUF:

Report ID: bp-675b859f-b15e-4482-922d-2df070200812 [@ gbm_bo_create ]
Submitted: 7 hours ago

Comment 5

[Ilgaz Öcal]

(In reply to Jan Andre Ikenmeyer [:darkspirit] from comment #4)

There is one crash report mentioning DMABUF:

Report ID: bp-675b859f-b15e-4482-922d-2df070200812 [@ gbm_bo_create ]
Submitted: 7 hours ago

It must be mine. Unfortunate thing is, the device is at my home and I have to remotely access it from work place. It is "frozen" again so I can't test anything further. I tested "mplayer" to make sure video acceleration works without crashing the OS, it indeed works.

Comment 6

[Martin Stránský [:stransky] (ni? me)]

(In reply to Jan Andre Ikenmeyer [:darkspirit] from comment #4)

There is one crash report mentioning DMABUF:

Report ID: bp-675b859f-b15e-4482-922d-2df070200812 [@ gbm_bo_create ]
Submitted: 7 hours ago

This crash means va-api is not used but SW decode to dmabuf surfaces is running. Perhaps the actual codec is not supported by the HW decoder.

Comment 7

[Darkspirit]

I assume I can't test this: My GTX 1060 (NV136) is "PureVideo HD VP8". Nouveau seems to support up to VP5.

(In reply to ilgaz from comment #0)

1. Does this h264 video look likewise broken for you?
2. Can you try to get a log of it? For example:
   $ pip3 install --upgrade mozregression
   $ MOZ_X11_EGL=1 MOZ_LOG="PlatformDecoderModule:5" ~/.local/bin/mozregression --launch 20200812034418 --pref gfx.webrender.all:true media.ffmpeg.vaapi.enabled:true media.ffvpx.enabled:false -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -P stdout > nouveau.txt

Comment 8

[Ilgaz Öcal]

Attachment: nouveau.txt

OpenGL: renderer: NVAC v: 3.3 Mesa 20.1.3 direct render: Yes 

(In reply to Jan Andre Ikenmeyer [:darkspirit] from comment #7)
> I assume I can't test this: My GTX 1060 (NV136) is "PureVideo HD VP8". Nouveau seems to support up to VP5.
> 
> (In reply to ilgaz from comment #0)
> 1. Does [this h264 video](https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605) look likewise broken for you?
Unfortunately yes

> 2. Can you try to get a log of it? For example:
> $ pip3 install --upgrade mozregression
> $ MOZ_X11_EGL=1 MOZ_LOG="PlatformDecoderModule:5" ~/.local/bin/mozregression --launch 20200812034418 --pref gfx.webrender.all:true media.ffmpeg.vaapi.enabled:true media.ffvpx.enabled:false -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 -P stdout > nouveau.txt

Comment 9

[Ilgaz Öcal]

output of vaainfo (it supports h2649
vainfo
vainfo: VA-API version: 1.8 (libva 2.7.1)
vainfo: Driver version: Mesa Gallium driver 20.1.3 for NVAC
vainfo: Supported profile and entrypoints
VAProfileMPEG2Simple : VAEntrypointVLD
VAProfileMPEG2Main : VAEntrypointVLD
VAProfileVC1Simple : VAEntrypointVLD
VAProfileVC1Main : VAEntrypointVLD
VAProfileVC1Advanced : VAEntrypointVLD
VAProfileH264ConstrainedBaseline: VAEntrypointVLD
VAProfileH264Main : VAEntrypointVLD
VAProfileH264High : VAEntrypointVLD
VAProfileNone : VAEntrypointVideoProc

Comment 10

[Martin Stránský [:stransky] (ni? me)]

The decode goes through VA-API correctly, I'm not sure why import fails. I'd need to check that on nvidia card then.

Comment 11

[Ilgaz Öcal]

I think everyone can handle some glitch&CPU load but the real issue is, if you let video play for a while it freezes the entire OS. Even control+alt+f keys doesn't work. A chip or nouveau bug? In 10-12 hours I can leave video playing and let the text file fill with data until OS freezes so you can understand what is happening. I thought it was impossible for usermode apps to freeze OS.

Comment 12

[Darkspirit]

(In reply to Martin Stránský [:stransky] from comment #10)

I'd need to check that on nvidia card then.

This one from April 2020 might be good to test:
https://www.phoronix.com/scan.php?page=news_item&px=Nouveau-K2020-Low-End-Card
~50-60€
https://www.heise.de/preisvergleich/asus-geforce-gt-710-90yv0e60-m0na00-a2286065.html?hloc=at&hloc=de

Chip GK208 "Kepler 2.0"

https://www.techpowerup.com/gpu-specs/nvidia-gk208.g572

PureVideo HD VP5

That's also the so far maximum supported VP version for VAAPI: https://nouveau.freedesktop.org/wiki/VideoAcceleration/

Comment 13

[Ilgaz Öcal]

Is it possible that firefox-vaapi/x11 triggers the same issue (freezing) mentioned here on a nvidia 9400?

G92 chips appear to hang when decoding H.264 videos. See this bug for more info. https://bugs.freedesktop.org/show_bug.cgi?id=82835

BTW, I wouldn't buy a nvidia card just to test this issue. I can arrange root access to Mac Mini if you wish.

Comment 14

[Ilgaz Öcal]

Testing VLC gave me error this time. Not sure it is related but pasting anyway:
VLC media player 3.0.11 Vetinari (revision 3.0.11-0-gdc0c5ced72)
[000055a5bbe7c670] main libvlc: Running vlc with the default interface. Use 'cvlc' to use vlc without interface.
[00007f3d20006c00] glconv_vaapi_x11 gl error: vaDeriveImage: operation failed
[00007f3d18e4f780] main video output error: video output creation failed
[00007f3d18c122b0] main decoder error: failed to create video output
[00007f3d18c122b0] avcodec decoder: Using G3DVL VDPAU Driver Shared Library version 1.0 for hardware decoding
[mpeg4 @ 0x7f3d18e2ef00] Failed setup for format vdpau: hwaccel initialisation returned error.
[00007f3d18c122b0] avcodec decoder error: existing hardware acceleration cannot be reused
[00007f3d4cc11bb0] glconv_vaapi_x11 gl error: vaDeriveImage: operation failed
[00007f3d4cc17aa0] glconv_vaapi_x11 gl error: vaDeriveImage: operation failed
[00007f3d18e7fbf0] main video output error: video output creation failed
[00007f3d18c122b0] main decoder error: failed to create video output
[mpeg4 @ 0x7f3d18e2ef00] get_buffer() failed
[mpeg4 @ 0x7f3d18e2ef00] thread_get_buffer() failed
[mpeg4 @ 0x7f3d18e2ef00] get_buffer() failed (-12 (nil))

Comment 15

[Ilgaz Öcal]

Using VLC I tested playing the testing video(In reply to Jan Andre Ikenmeyer [:darkspirit] from comment #7)

I assume I can't test this: My GTX 1060 (NV136) is "PureVideo HD VP8". Nouveau seems to support up to VP5.

(In reply to ilgaz from comment #0)

1. Does this h264 video look likewise broken for you?

I also tested playing this video on VLC. it didn't have any visible issues but it seems vaapi fails.

[ilgaz@ilgaz-macmini31 firefox]$ mv attachment.cgi?id=9149605 test.mp4
[ilgaz@ilgaz-macmini31 firefox]$ vlc test.mp4
VLC media player 3.0.11 Vetinari (revision 3.0.11-0-gdc0c5ced72)
[0000564eb65c1670] main libvlc: Running vlc with the default interface. Use 'cvlc' to use vlc without interface.
[00007f34dc006c00] glconv_vaapi_x11 gl error: vaDeriveImage: operation failed
[00007f3510000d20] main video output error: video output creation failed
[0000564eb72e6a20] main decoder error: failed to create video output
[0000564eb72e6a20] avcodec decoder: Using G3DVL VDPAU Driver Shared Library version 1.0 for hardware decoding

Comment 16

[Darkspirit]

https://nouveau.freedesktop.org/wiki/VideoAcceleration/ > Firmware:
Do you have https://aur.archlinux.org/packages/nouveau-fw installed?

Comment 17

[Ilgaz Öcal]

(In reply to Jan Andre Ikenmeyer [:darkspirit] from comment #16)

https://nouveau.freedesktop.org/wiki/VideoAcceleration/ > Firmware:
Do you have https://aur.archlinux.org/packages/nouveau-fw installed?

Yes, double checked.

Comment 18

[Martin Stránský [:stransky] (ni? me)]

(In reply to ilgaz from comment #15)

[ilgaz@ilgaz-macmini31 firefox]$ mv attachment.cgi?id=9149605 test.mp4
[ilgaz@ilgaz-macmini31 firefox]$ vlc test.mp4
VLC media player 3.0.11 Vetinari (revision 3.0.11-0-gdc0c5ced72)
[0000564eb65c1670] main libvlc: Running vlc with the default interface. Use 'cvlc' to use vlc without interface.
[00007f34dc006c00] glconv_vaapi_x11 gl error: vaDeriveImage: operation failed
[00007f3510000d20] main video output error: video output creation failed
[0000564eb72e6a20] main decoder error: failed to create video output
[0000564eb72e6a20] avcodec decoder: Using G3DVL VDPAU Driver Shared Library version 1.0 for hardware decoding

Looks like dmabuf surface export fails here on the driver/mesa side.

Comment 19

[Martin Stránský [:stransky] (ni? me)]

We can't fix the failure but we should switch SW decode (Bug 1724385).