Skip to content

L
libva
An error occurred fetching the project authors.
Switch branch/tag
  1. 03 Sep, 2009 1 commit
  3. 28 Aug, 2009 1 commit
  5. 24 Aug, 2009 1 commit
  7. 14 Aug, 2009 1 commit
  9. 08 Jul, 2009 3 commits
  11. 06 Jul, 2009 1 commit
  13. 03 Jul, 2009 1 commit
  15. 01 Jul, 2009 1 commit
  17. 02 Jul, 2009 2 commits
  19. 25 Jun, 2009 1 commit
  21. 24 Jun, 2009 1 commit
  23. 13 Jun, 2009 1 commit
  25. 12 Jun, 2009 1 commit
  27. 03 Jun, 2009 1 commit
    • Austin Yuan's avatar
      Remove uncessary link form libva.so by adding flag RTLD_NODELETE into · 74047bc1
      Austin Yuan authored 
      dlopen

Here is the background of this fix
Investigation indicates MRST Moblin Alpha1 0529 and 0520 uses the same libva source package
(which is from graphics Alpha1.4 pakcage
\\mid-depot.amr.corp.intel.com\Exchange\Moblin2\PackageSubmit\1.0.7_05142009-4_Alpha1.4),
and the rootcause is that the linkage of libva dependence libraries is
chaned in 0529 build. I am not sure why 0529 build has this change, and this
change indeed causes libva application segment fault issue.

See the detailed investigation and explanation in the attached
libva-0520-vs-0529.PNG picture.

Reproduce Steps(steps,current result, reproduce possibility)
===========================================================
(1) boot 0529 build
(2) Install libva testsuits into 0529 build
(3) run "mpeg4vld -x -i /var/clips/demo.m4v"

Expected result:
===========================================================
libVA application should exit cleanly

Possible root cause:
================================
There was a global change which helps reduce unnecessary linking
utilizing a
feature in binutils. We could opt liva out of this easily, however I
would
recommend fixing the package to do the right linking instead of relying
on the
tools to do so, Arjan, any suggestions?

Anas
=============================
summary of the picture:

the "new" libva does no longer link to

* libXv

* libdrm

* librt

* libpthread

HOWEVER; libva doesn't USE any of these! I don't see how not linking to
these could lead to a crash, since they truely are not used.

================================
Basically it is a known issue. If we remove these library link from
libVA, we will always get a segment fault when XCloseDisplay is called in
application.
Libva doesn't use these liXv/libdrm libraries, but libVA will dlopen HW
specific driver, and the driver links with these libXv/libdrm libraries.
We found the issue can be worked around by adding the link into libVA.
Using binutil to remove unnecessary link makes sense for most of libraries,
but for libva, it disables our workaround.

Austin
Signed-off-by: default avatarAustin Yuan <shengquan.yuan@intel.com>
      74047bc1
  29. 12 May, 2009 1 commit
  31. 29 Apr, 2009 1 commit
  33. 22 Apr, 2009 2 commits
  35. 26 Mar, 2009 1 commit
  37. 25 Mar, 2009 5 commits
  39. 02 Mar, 2009 1 commit
  41. 21 Jan, 2009 1 commit
  43. 12 Jan, 2009 1 commit
    • Austin Yuan's avatar
      Apply the patch to split va and display/x11 from Gwenole Beauchesne... · e1b1327b
      Austin Yuan authored 
      Apply the patch to split va and display/x11 from Gwenole Beauchesne [mailto:gbeauchesne@splitted-desktop.com]

Bellow is his explanation:

> Finally, looking further at <va_x11.h>, I think it should be enough to have
> vaInitialize() in display-dependent headers/libs. The va_x11_getDriverName()
> suggestion was to factor out the thing at the implementation (source
> code/files) level.
>
> Or we could keep vaInitialize() in common lib and rather have vaGetDisplay()
> in the display-specific part? And, while being at it, also rename the
> function to vaCreateDisplay(), to be meaningful about the API change?
>
> Besides, for a different windowing system, we probably would need more than
> just the Display (as we have in X11 land) anyway. e.g. what about OpenGL,
> OpenGL E|S? I don't know, it's just an idea.
>
> I read that Canmore/Sodaville are using the same engines as the Poulsbo
> (SGX535 and VXD370). However, the former platforms only support OpenGL E|S.
> So, how does video acceleration work here? I know it works, I saw it but
> since we still haven't received the machines, I just don't know about the
> actual API. You'd probably want libVA there too.
>
> Splitting libVA between a Core API and a Display API would make it possible
> to reduce code duplication from a player point of view. i.e. I don't think
> it's necessary to have client applications implement
> vaBeginPicture()..vaEndPicture() sequences themselves. I think it should be
> the role of the codec library (ffmpeg, in my case), and it should be able to
> do so without an explicit dependency on X11.
>
> On the other hand, the Core API won't be useful/functional alone. So, that
> could be confusing too.
>
> In practise, I would like to have it working as follows. It's my ideal
> vision, not necessarily the right/correct one. ;-)
>
> Roles of a codec implementation library:
> - Create buffers
> - Render the pictures, in the vaBeginPicture()..vaEndPicture(),
> vaRenderPicture() sense
>
> Roles of a player application:
> - Create display, surfaces, and decode pipeline for the codec library
> - Render the picture, visually, i.e. in the vaPutSurface() sense
>
> Example use:
> VApplication|initialize display
> CodecLibrary|characterise bitstream (codec and other useful info)
> VApplication|create decode pipeline
> VApplication|create surfaces
> CodecLibrary|create buffers	(1)
> CodecLibrary|render picture	(2)
> VApplication|display picture	(3)
> repeat (1) -> (3) while the end of stream is not reached
> VApplication|destroy everything
>
> Have CodecLibrary linked against libva-core-VERSION.so.MAJOR, without any
> dependency on windowing system library.
>
> Have VApplication linked against libva-x11-VERSION.so.MAJOR, itself linked
> against libva-core-VERSION.so.MAJOR and other windowing system libraries.
>
> Regards,
> Gwenole.
>
Signed-off-by: default avatarAustin Yuan <shengquan.yuan@intel.com>
      e1b1327b
  45. 29 Aug, 2008 1 commit
  47. 29 Apr, 2008 1 commit
  49. 14 Mar, 2008 1 commit
  51. 07 Feb, 2008 1 commit
  53. 17 Dec, 2007 1 commit
  55. 07 Dec, 2007 1 commit
  57. 07 Nov, 2007 1 commit
  59. 31 Oct, 2007 1 commit
  61. 25 Sep, 2007 1 commit
  63. 20 Sep, 2007 1 commit