[FFmpeg-devel] need help fixing sws on PPC
Sun Oct 5 21:27:31 CEST 2008
On Wed, Sep 10, 2008 at 01:42:44PM +0200, Vitor Sessak wrote:
> Michael Niedermayer escreveu:
> > Hi
> > Can someone who has a PPC and a x86 play a little with swscale to find
> > out which convertions do not match?
> > This is rather important as our regression tests as well as fate depend
> > on it.
> > Note -sws_flags +bitexact should be used also -sws_flags +accurate_rnd
> > may be needed.
> > Also note, some scalers still use floats to find coeffs (like spline)
> > iam not interrested in them, that is currently area, nearest neighbor
> > bilinear and bicubic are the onlx ones iam interrested in.
> > And last, if something does not match, pleas explain how it does differ
> > +-1 rounding or red/blue exchanged or ...
> Besides rounding, there is still something wierd about swscale. For
> example, I was looking at he idcin test
> http://fate.multimedia.cx/index.php?test_spec=107 .
> Unless I am missing something, there should be no calculation (and even
> less rounding) doing PAL8 -> RGB conversion, but it still gives
> different output with and without swscaler.
I did mention several times that swscale uses yuv internally, the old
scaler only supports yuv420 scaling. So which ever you use when scaling
is involved pal->rgb will do rgb->yuv->rgb as well, and with the old one
chroma would be subsampled by 2x2.
if you only care about the unscaled pal->rgb24/32 case, you can add a special
converter for that case, which should be trivial.
if you want the scaled case to be more exact, the input pal->yuv code and
horizontal scaler should be where you should look, making it use more
than 8 bit per component yuv.
Adding a rgb path to the scaler is possible as well but wont make much
of a difference as the 8bit rgb output will be much more limiting than
any rounding 16bit yuv could introduce.
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
Democracy is the form of government in which you can choose your dictator
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 189 bytes
Desc: Digital signature
More information about the ffmpeg-devel