[FFmpeg-devel] [PATCH] hardcode motionpixels.c mp_rgb_yuv_table

[Reimar Döffinger]

On Sun, Nov 01, 2009 at 06:50:14PM +0100, Michael Niedermayer wrote:
> On Wed, Oct 28, 2009 at 08:30:01PM +0100, Reimar D?ffinger wrote:
> > This would be interesting since using a matrix-vector-multiplication it
> > seem only to be possible to come within +-3 of the values in this table,
> > so if it has to be exact there is not really a (decently performant)
> > way to avoid the table completely.
> 
> The table could be made smaller ...

Yes, though it seemed not worth it to me.

> also i dont think you have a formal proof that there is no decently
> performant solution withot a big table :)

Aw, don't read what I wrote but what I meant :-P.
I actually haven't tried anything beyond simple linear regression,
which was the limit of what I was ready to try.