[MPlayer-G2-dev] libvo2 documentation and pre22 iz out

[D Richard Felker III]

On Thu, May 15, 2003 at 10:50:08AM +0200, Alban Bedel wrote:
> Hi D Richard Felker III,
> 
> on Wed, 14 May 2003 16:05:26 -0400 you wrote:
> 
> > At some point, we really need to work out a *good* architecture for
> > all the *_vid drivers to be in kernelspace, with access to dma, irq,
> > etc. through the drm (if available)...
> 
> Perhaps but as the *_vid drivers are intended for high performance
> drivers it really depend on the way the card work. With the banshee
> for example 'internal blit' (ie copy inside the card mem) are
> essential as they can do scaleing and/or colorspace conversion.
> Moreover for this card the overlay can only do up scaling so when
> you want to downscale you have to either don't use the overlay at
> all or downscale when copying to the overlay mem and resetup
> the overlay source size.
> So i doubt that we can really get every bit of performance out of
> every card with a general approch.
> Perhaps that most modern cards can be well handled in general way
> but i doubt we'll find an interface wich will suit every kind of
> card with scaleing and/or colorspace conversion caps.

As far as I can tell, all we need so far is a CAN_DOWNSCALE flag and a
list of supported colorspaces... Doesn't sound too hard. VO drivers
are already required to identify whether or not they can downscale, so
the vo driver could just return the flag from the *_vid driver.

As for non-overlay-based scalers, I think they're best kept separate.
I was talking about a general driver architecture for video overlays
(like vidix was supposed to be, except non-broken), not for all video
scaling and colorspace conversion stuff.

Rich