[MPlayer-dev-eng] [PATCH] libass support for VDPAU

[Grigori G]

On Fri, 6 Feb 2009 23:01:58 +0100, Reimar Döffinger
<Reimar.Doeffinger at stud.uni-karlsruhe.de> wrote:
> 
> I don't know how much point there is in reviewing this while vdpau is
> not yet in, but how is the speed of this?
>

I'm not sure either... :)

> Having many hundreds of tiny textures cause serious performance issues
> with -vo gl.
>

And it's the same here, basically. The code tries to reuse bitmapsurfaces
as much as possible though.
The real problem is how libass rendering works, software-rendering to a
single RGBA bitmap would probably be faster and avoid a few of the
problems libass has at the moment.

>> +#define EOSD_MAX_SURFACES 1000
> 
> There should not be a fixed limit.
>

Acknowledged, this should be implemented with a singly-linked list or
similar.

>> +            ax = round_pot(i->w);
>> +            ay = round_pot(i->h);
> 
> Why? I expected the hardware all supports non-power-of-two since a long
> time.
> If you really _have_ to use power-of-two, be very careful that there are
> no issues due to the not-initialized parts of the surface.
> That's all for a first quick review.
>

VDPAU supports non power-of-two bitmapsurfaces, but making them
power-of-two
makes it possible to reuse more of them whenever something on screen
changes.
The compositing is only done on the area actually used by the bitmap.

Grigori