[PATCH 3/4] drm: Add drm_format_{horz, vert}_chroma_subsampling() utility functions

Ville Syrjälä ville.syrjala at linux.intel.com
Fri Apr 13 05:09:04 PDT 2012

On Fri, Apr 13, 2012 at 04:42:04AM +0200, Roland Scheidegger wrote:
> Am 12.04.2012 14:23, schrieb Ville Syrjälä:
> > On Wed, Apr 11, 2012 at 08:53:08PM +0200, Roland Scheidegger wrote:
> >> Am 05.04.2012 20:35, schrieb ville.syrjala at linux.intel.com:
> >>> From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> >>>
> >>> These functions return the chroma subsampling factors for the specified
> >>> pixel format.
> >> Hmm not really related but looking at it this reminds me these formats
> >> always look a bit underspecified wrt chroma subsample positions. Are
> >> these fixed, undefined, or what (even mpeg1 and mpeg2 differ there, and
> >> let's not even talk about the big mess that interlaced mpeg2 is)?
> > 
> > My thinking is that the sample positions can be configured through
> > some other means. It doesn't affect the in memory layout at all so
> > tying it to the format isn't necessary.
> > 
> > This also depends on the hardware. On some hardware you can configure
> > it quite freely, but on some you may be left with just the one option,
> > which is often not even documented.
> > 
> > I sent another patch [1] some time ago which provides a helper function
> > for drivers dealing with good hardware.
> > 
> > [1] http://lists.freedesktop.org/archives/dri-devel/2011-December/017585.html
> > 
> I guess that makes sense. As long as those "other means" are available.
> Not like with xv where I think drivers just default to what is probably
> the most likely subsample positions, so assuming mpeg2/4 progressive or
> something (which gets mpeg1 slightly wrong), if they even bother to care
> about such subtleties :-).

Yeah. I think a plane property is the best choice here. If the hardware
doesn't allow you to configure this stuff, the property would just
indicate the fixed configuration that the hardware supports. But this
is again another case where we really want standardised and well
defined properties.

Ville Syrjälä
Intel OTC

More information about the dri-devel mailing list