[Mesa-dev] Status of VDPAU and XvMC state-trackers (was Re: Build error on current xvmc-r600 pipe-video)

[Christian König]

Am Montag, den 09.05.2011, 20:04 +0100 schrieb Andy Furniss:
> Andy Furniss wrote:
> 
> > dd6cd206a6395be651bc965580e17c0d63513c7b
> >
> > [g3dvl] correctly implement non power of two buffers
> >
> > regresses rendering on my RV670.
Very interesting, sounds like a bug in the shader compiler for R6xx, but
this is not prio one on my todo list.

> I see with the weekend changes and master merge that the "new" xvmc 
> artifacts on my rv790 are now gone without patching.
Yes Alex patches combined with mine seems to indeed work right.

> mplayer vanilla is working and vdpau perf is much better.
Except for subpictures, a couple of crashes and a bunch of other still
open problems...

> On my rv790 there are still vdpau decode artifacts that don't show with 
> xvmc -
I tried to compare the shader input and output between xvmc and vdpau
for the last couple of days now, but still haven't found all the bugs
yet, but this should just be a matter of time.

> My rv670 is still suffering quite different problems both xvmc and vdpau -
Please try the following patch with xvmc:
--- a/src/gallium/drivers/r600/r600_video_context.c
+++ b/src/gallium/drivers/r600/r600_video_context.c
@@ -40,5 +40,5 @@ r600_video_create(struct pipe_screen *screen, void
*priv)
    if (!pipe)
       return NULL;
 
-   return vl_create_context(pipe, false);
+   return vl_create_context(pipe, true);
 }

It won't work with vdpau, because it probably causes a buffer overflow
and/or crash right now.

Christian.