master has issues with video replay (mesa related?)

  • Hello,


    it seems that iMX6 does currently not build, since a patch for etnaviv does not apply properly. Not sure if this is the right place to report here....

    Thanks,

    Rainer

  • I've pinged the maintainer. The workaround will be to create another $PROJECT entry in packages/linux/package.mk so that NXP project is compiled using an eariler kernel that the patches were generated agaisnt. Until a proper kernel bump is done.

  • I saw the recent commits, which resolve this issue. Now the build fails later on


  • ^ that probably fixes it

  • The image boots on an Cubox-i/imx6. Basic operations works, slideshows work.


    Playing videos does not work. There seems to be an audio issue, but the main issue is that the TV reports that there is no HDMI source anymore.


    In kodi.log, I see (full log at kodi.log video starts around 16:53)


    Code
    1. 2020-03-21 16:59:22.643 T:988 ERROR: CAESinkALSA - snd_pcm_writei(-32) Broken pipe - trying to recover
    2. 2020-03-21 16:59:22.646 T:980 ERROR: CDRMAtomic::DrmAtomicCommit - test commit failed: Numerical result out of range
    3. 2020-03-21 16:59:22.797 T:988 ERROR: Previous line repeats 3 times.
    4. 2020-03-21 16:59:22.797 T:988 ERROR: CAESinkALSA - snd_pcm_writei(-32) Broken pipe - trying to recover
    5. 2020-03-21 16:59:22.812 T:980 ERROR: CDRMAtomic::DrmAtomicCommit - test commit failed: Numerical result out of range
    6. 2020-03-21 16:59:22.950 T:988 ERROR: Previous line repeats 3 times.
    7. 2020-03-21 16:59:22.950 T:988 ERROR: CAESinkALSA - snd_pcm_writei(-32) Broken pipe - trying to recover
    8. 2020-03-21 16:59:22.979 T:980 ERROR: CDRMAtomic::DrmAtomicCommit - test commit failed: Numerical result out of range
    9. 2020-03-21 16:59:23.103 T:988 ERROR: Previous line repeats 2 times.


    and dmesg is full of



    entries. This is not the first one though, the first one aged out of the history already.


    Here is in addition a dmesg output after a reboot:


    dmesg.log

    Edited once, last by rdorsch ().

  • You will need to use egl rendering for video playback but that will only work at about 15FPS. We cannot reliably do direct to plane rendering on imx6 because of restrictions with the planes.

  • How is this related to the 1080p50 mentioned in #kodi ?


    [22:10] <TeamKodi> [unknown] decoder is rated for 1080p30 only (we nearly got 1080p50 going)

    [22:10] <TeamKodi> [unknown] for etnaviv + coda it's in a very good shape. dma_buf import works (for egl) or even direct yuv scanout

    [22:11] <TeamKodi> [unknown] why no one works on it anymore: it's just 1080p capable

    [22:12] <TeamKodi> [unknown] it should be in quite good shape though with v4l2 decoder

    [22:14] <TeamKodi> [unknown] And: no one has time anymore ;)


    And how do I use egl rendering? Are the 15 fps for 1080p will it increase for 720p or get even worse?

  • EGL rendering is set in settings -> media playback -> video -> DRM prime render method.


    And yes, you should be able to get nearly full speed 720p

  • Thanks, Lukas


    I found the setting in


    settings -> player -> media playback -> video -> PRIME render method


    and put it to EGL as you recommended. Then I had to change the audio device in system settings, the default used spif instead of HDMI and enabled passthrough.


    After that it worked amazingly well.


    I saw two issues


    • the coloring of the videos was wrong. They have been almost black and white, but some colors were mixed in, sometimes they seem to look unrelated to the video.
    • I can always play a video once then it does not start anymore


    Full log is here.

  • Hi rdorsch,


    I'm new on the forum and was following this interesting thread.

    I'm currently facing the same problem with the "wrong colours of the videos". I quickly tried to apply the 2 patches stacks from your links.

    Unfortunately the 1st one seems really old and some source files don't exist anymore (i.e. etnaviv_2d.c,etnaviv_2d.h). Regarding the second stack I edited the patches to fit in the source code but even if it compiles properly the results is worse --> the video is made of multiple tiles in the wrong order and with wrong colours. I may have done something wrong though...


    I was wondering if you made some progress on your side or if you have more information to share ?


    Thank you


    (I'm currently testing on a wandboard-iMX6q)