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

[d2kx]

After using Kano's install script, running mplayer -vo vaapi -va vaapi <filename> gives the following error, and there's no video displayed (just audio).

libva: libva version 0.31.0-sds3
libva: va_getDriverName() returns 0
libva: Trying to open /usr/local/lib/va/drivers/fglrx_drv_video.so
libva: va_openDriver() returns -1
[vo_vaapi] vaInitialize(): unknown libva error
Error opening/initializing the selected video_out (-vo) device.

I expect this is because I'm using the stream beta driver, but nobody else mentioned it so I thought I would

I'm using ubuntu 9.10 with an ati 5750 and the catalyst 9-10 driver won't load as it doesn't recognise my card.

Any ideas/hope?

Thanks, Evan.

It's "-vo vaapi:gl -va vaapi" (you missed the :gl) for now.

[gbeauche]

What exactly does that mean? Will MPEG2/4 see any benefit from XvBA in the near term or at all? Most of my use (>95% by percentage of play time) is MPEG2.

The UVD only supports H.264 and VC-1 at VLD level. MPEG-2 is supported but at the IDCT level. Supporting other codecs would require offload to the GPU through shader programs.

Back to the code itself... Is any effort going into integrating it into project mainlines - libva, FFMPEG, Mplayer, MythTV, VLC etc? I realize acceptance is project dependant but are the sources being marketed to them or is it destined to stay in the SDS tree for some long closed development cycle.

Sources to what? The sources to xvba-video will remain closed as long as the XvBA API is private. Besides, the public version of xvba-video has reduced features wrt. private versions we develop for our own (embedded) solutions. We don't have the sources to the XvBA driver though I would be very happy to have and everybody knows it...

[gbeauche]

After using Kano's install script, running mplayer -vo vaapi -va vaapi <filename> gives the following error, and there's no video displayed (just audio).

libva: libva version 0.31.0-sds3
libva: va_getDriverName() returns 0
libva: Trying to open /usr/local/lib/va/drivers/fglrx_drv_video.so
libva: va_openDriver() returns -1
[vo_vaapi] vaInitialize(): unknown libva error

This means fglrx_drv_video.so was not found in /usr/local/lib/va/drivers/ or is not the right binary (32-/64-bit mismatch) -- please check your installation.

[Vighy]

an what about:

ben@Obi-Wan ~ $ vainfo 
libva: libva version 0.31.0-sds3
libva: va_getDriverName() returns 0
libva: Trying to open /usr/lib64/va/drivers/fglrx_drv_video.so
libva error: dlopen of /usr/lib64/va/drivers/fglrx_drv_video.so failed: /usr/lib64/va/drivers/fglrx_drv_video.so: wrong ELF class: ELFCLASS32
libva: va_openDriver() returns -1
vaInitialize failed with error code -1 (unknown libva error),exit

what does wrong ELF class: ELFCLASS32 mean?

[Vighy]

an what about:

ben@Obi-Wan ~ $ vainfo 
libva: libva version 0.31.0-sds3
libva: va_getDriverName() returns 0
libva: Trying to open /usr/lib64/va/drivers/fglrx_drv_video.so
libva error: dlopen of /usr/lib64/va/drivers/fglrx_drv_video.so failed: /usr/lib64/va/drivers/fglrx_drv_video.so: wrong ELF class: ELFCLASS32
libva: va_openDriver() returns -1
vaInitialize failed with error code -1 (unknown libva error),exit

what does wrong ELF class: ELFCLASS32 mean?

ah! I found: it's the ebuild that just installs 32bit binary of xvba-video! ahh!

[gbeauche]

an what about:

ben@Obi-Wan ~ $ vainfo 
libva: libva version 0.31.0-sds3
libva: va_getDriverName() returns 0
libva: Trying to open /usr/lib64/va/drivers/fglrx_drv_video.so
libva error: dlopen of /usr/lib64/va/drivers/fglrx_drv_video.so failed: /usr/lib64/va/drivers/fglrx_drv_video.so: wrong ELF class: ELFCLASS32
libva: va_openDriver() returns -1
vaInitialize failed with error code -1 (unknown libva error),exit

what does wrong ELF class: ELFCLASS32 mean?

fglrx_drv_video.so is a 32-bit binary whereas the 64-bit one is expected. BTW, I never really tested the 64-bit binaries, so it does interest me to know whether they work or not.

[rjwaldren]

The UVD only supports H.264 and VC-1 at VLD level. MPEG-2 is supported but at the IDCT level. Supporting other codecs would require offload to the GPU through shader programs.

Sources to what?

Since mplayer uses the XvBA library for XvMC I assumed that passing MPEG2 to hardware more closely tied to the the recent developments than it is - I'm no developer though. As for the sources - I was more interested in what if any response there is to submission of your Mplayer, libva patches upstream. I know you won't be releasing the driver sources and we'll have to wait through the trickle down.

I do plan on testing what the new driver can do though.

Bridgman, XvMC works in mplayer but the performance is only marginally better than Xv in my experience and still has the Xv washout. GL beats them both for playback quality, IMO, and is roughly on par for CPU usage. <-All use too much

I'll probably just get a cheap NV9xxx card (along with it's hassles) until ATI vid accel support matures on the OSS side. I always give ATI the first shot... But my NV6100IGP generally is less of a hassle than my HD4850 under linux. The power of the HD4850 is much greater but when it comes to 1080p video or 1080i OTA streams they both suck. It's a great card under win7 though.

[linuxjacques]

fglrx_drv_video.so is a 32-bit binary whereas the 64-bit one is expected. BTW, I never really tested the 64-bit binaries, so it does interest me to know whether they work or not.

I'm on 64-bit Fedora 11 with Radeon HD 2400 XT and I get:

vainfo
libva: libva version 0.31.0-sds3
libva: va_getDriverName() returns 0
libva: Trying to open /usr/local/lib/va/drivers/fglrx_drv_video.so
libva: va_openDriver() returns 0
vainfo: VA API version: 0.31
vainfo: Driver version: Splitted-Desktop Systems XvBA backend for VA API - 0.5.1
vainfo: Supported profile and entrypoints
xvba_video: XVBA_CreateContext(): status 11
xvba_video: XVBA_CreateContext(): status 11
xvba_video: XVBA_CreateContext(): status 11
X Error of failed request:  BadLength (poly request too large or internal Xlib length error)
  Major opcode of failed request:  35 (X_AllowEvents)
  Serial number of failed request:  14
  Current serial number in output stream:  14

[Vighy]

fglrx_drv_video.so is a 32-bit binary whereas the 64-bit one is expected. BTW, I never really tested the 64-bit binaries, so it does interest me to know whether they work or not.

no they don't!

mplayer crashes 1-2 sec after launched.
And while trying to take a snapshot of what shows in that window for a while (because it shows something scrambled) it crashed my Xorg

the error it gives is:

xvba_video: XVBA_CreateDecode(): status 2
[vo_vaapi] vaCopySurfaceGLX(): unknown libva error
xvba_video: XVBA_CreateDecode(): status 2
[vo_vaapi] vaCopySurfaceGLX(): unknown libva error
xvba_video: error: Assertion failed in file xvba_decode.c at line 774

vainfo (just in case):

vainfo 
libva: libva version 0.31.0-sds3
libva: va_getDriverName() returns 0
libva: Trying to open /usr/lib64/va/drivers/fglrx_drv_video.so
libva: va_openDriver() returns 0
vainfo: VA API version: 0.31
vainfo: Driver version: Splitted-Desktop Systems XvBA backend for VA API - 0.5.1
vainfo: Supported profile and entrypoints
      VAProfileMPEG2Simple            :	VAEntrypointIDCT
      VAProfileMPEG2Main              :	VAEntrypointIDCT
      VAProfileH264High               :	VAEntrypointVLD
      VAProfileVC1Advanced            :	VAEntrypointVLD

and libva is the last (4 nov 2009) 0.31.0+1-sds7

if you need something, just tell me.

[bridgman]

Bridgman, XvMC works in mplayer but the performance is only marginally better than Xv in my experience and still has the Xv washout.

Yeah, the consensus here a year or so ago seemed to be that with modern CPUs there wasn't really much need for MPEG2 acceleration anyways. It's really the combination of high bitrate H.264/VC-1 and single-threaded software decoders that makes hardware decode acceleration so attractive right now