Announcement

Collapse
No announcement yet.

AMD's UVD2-based XvBA Finally Does Something On Linux

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Originally posted by bugmenot View Post
    i installed catalyst 10.1 on Ubuntu 9.10
    download libva libva-dev xvba-video packages from http://www.splitted-desktop.com/~gbeauchesne/ and install it.
    then download mplayer-vaapi-latest.tar.bz2 too.
    apt-get buil-dep mplayer
    ./checkout-patch-build.sh
    script successfully download and compiled whle mplayer.
    when i try play file ./mplayer some-file.mp4 it work.
    when i try run ./mplayer -vo vaapi -va vaapi some-file.mp4 i get this

    CPU usage is 0-1% so acceleration work
    add -vo vaapi:gl[:reflect] is same.

    using only -vo vaapi withou -va vaapi lead to normal output only colors is switched. people have blue faces. and CPU usage is 20-30%

    using only -va vaapi is normal but CPU is still 20-30%
    Reporting, that I got very similar results on my HD5650 mobility. ( http://imagebin.org/83738 )
    So I guess there's no way to use the gl acceleration with the 5xxx series yet, no xorg.conf tricks might help?

    Comment


    • Originally posted by krionius View Post
      Reporting, that I got very similar results on my HD5650 mobility. ( http://imagebin.org/83738 )
      So I guess there's no way to use the gl acceleration with the 5xxx series yet, no xorg.conf tricks might help?
      I will let Kano translate my words in something more explicit but short answer is 'no'. However, the bug is simple to fix for AMD since the same happened for 4xxx series some time ago. You might try to binary patch the library and enable the workarounds for 5xxx series too? The PCI IDs are hardcoded with x86 mov, cmp instructions.

      Anyway, you are stuck or you could try to find an NVIDIA MXM module and check it's compatible with your laptop. Do you have an IGP in there?

      Comment


      • Originally posted by gbeauche View Post
        I will let Kano translate my words in something more explicit but short answer is 'no'. However, the bug is simple to fix for AMD since the same happened for 4xxx series some time ago. You might try to binary patch the library and enable the workarounds for 5xxx series too? The PCI IDs are hardcoded with x86 mov, cmp instructions.

        Anyway, you are stuck or you could try to find an NVIDIA MXM module and check it's compatible with your laptop. Do you have an IGP in there?
        Unfotunately no IGP enabled in this model AFAIK (no hint in BIOS and lspci shows only the ATI card, so it's not connected into the big picture ), so I have to cope with the ATi.

        About the PCI IDs, i'm open to experimentation. So theoretically it could work, just the PCI IDs are mixed up/missing, or something else? I would be happy to try it if someone has some info about this - i have a hex editor if that's what I need

        Comment


        • Just try multithreaded decode:

          http://kanotix.com/files/fix/mplayer...t-snapshot.txt

          With a multicore cpu this could help you. In my experience video players with pure vdpau support are best (just require nv hardware) - mplayer for lots of files just no m2ts with subtitles (i really wait for that) or xbmc for m2ts with subtitles. vlc only works with a wrapper, ati + vlc = impossible, nvidia + vlc = stupid green line below video. subtitles work with m2ts - that's what i mostly watch...

          Comment


          • well, the mplayer/vlc i tried already uses multiple cores (i have an i5 CPU in notebook), so it plays well 1080p too.

            But I presume that the GPU eats less power while decoding it. Am I right?

            Also the point I want to get to is: is it possible to hack the binary to workaround this bug? If so that would be quite helpful for all those wanting to use this feature on 5XXX, and can't wait for AMD to fix this in a later release.

            So hints and info about this would be helpful! What kind of workaround worked for the 4xxx?

            Comment


            • Originally posted by gbeauche View Post
              I will let Kano translate my words in something more explicit but short answer is 'no'. However, the bug is simple to fix for AMD since the same happened for 4xxx series some time ago. You might try to binary patch the library and enable the workarounds for 5xxx series too? The PCI IDs are hardcoded with x86 mov, cmp instructions.
              Okay, i've been thinking about this. So the card's ID enables a Workaround in one of the libraries? Can you tell me which library it is, and where can i find some PCI ids for those 4xxx cards that uses the workaround? I will try to look up them in the binary and replace with my cards' PCI id.

              Comment


              • xvba-video 0.6.5

                A new version of xvba-video, the XvBA backend to VA-API, is now available at:
                http://www.splitted-desktop.com/~gbe...ne/xvba-video/

                Version 0.6.5 - 08.Feb.2010
                * Add brightness/contrast/hue/saturation display attributes
                * Fix vaPutSurface() window resize. e.g. when switching to full-screen mode
                * Allow vaPutSurface() to render to multiple drawables from a single surface

                Notes:
                - My ProcAmp adjustments are probably not fully correct. e.g. hue doesn't preserve luminance yet. Besides, this uses an extra FBO.
                - The last change workarounds a bug in the driver and now makes it possible to use VA-API acceleration with Gnash with the the AGG renderer. However, this exhausts another performance problem (flickering in windowed mode) of the driver. You can workaround that with XVBA_VIDEO_PUTSURFACE_FAST set to "yes" or "1". The semantics are not fully equivalent and can cause problems, hence it's disabled by default though it's designed to work with Gnash and MPlayer.

                Comment


                • With vlc it is still just green and shows lots of XVBA_GetSurface (status 2) errors. When do you want to fix it? Also the stupid green line with nvidia+vlc should be removed.

                  Comment


                  • Originally posted by Kano View Post
                    With vlc it is still just green and shows lots of XVBA_GetSurface (status 2) errors. When do you want to fix it? Also the stupid green line with nvidia+vlc should be removed.
                    Probably when you will be capable enough to produce correct bug reports that can be reproduced? Besides, I don't maintain VLC, did you know?

                    Comment


                    • My reports are 100% correct. I mailed you screenshots already.

                      Comment

                      Working...
                      X