[MPlayer-users] uspp filter - colorspace

Reimar Döffinger Reimar.Doeffinger at gmx.de
Sat Mar 10 20:24:51 CET 2012


On Sat, Mar 10, 2012 at 07:25:59PM +0100, chronosek wrote:
> W dniu 2012-03-10 19:09, Reimar Döffinger pisze:
> >On Thu, Mar 08, 2012 at 07:33:08PM +0100, chronosek wrote:
> >>PS2. i wonder if someone could speed up that so good filter
> >>(i use normally avisynth filters (with rgb24 colorspace) but use
> >>mencoder too - only just for that filter)
> >Is is based on the snow encoder, so anything speeding that up will help.
> >One easy way is to add a way to disable the call to encode_subband
> >in it (should be about 20% faster).
> >_______________________________________________
> >MPlayer-users mailing list
> >MPlayer-users at mplayerhq.hu
> >https://lists.mplayerhq.hu/mailman/listinfo/mplayer-users
> >
> But maybe can use more threads ? (i have 6 cores all with ht , but
> uspp uses only one thread....) maybe some ssse3 or sse4 operations
> can speed that snow encoder?

Threading: difficult and IMHO not the right thing to do at the stage the
decoder is, see next point.
SSSE3 etc.: Just asm-optimizing the core functions at all would be
a big help, so far it is all pure C.
I wouldn't be surprised if one could double the speed with that alone.
Don't expect me to do it, this is purely a FFmpeg thing anyway and
doesn't really have anything to do with MPlayer directly.
I would say the same applies to RGB or 4:4:4 support, the snow encoder
would have to support that first.

> even 20% is good  (with uspp=6:8 i have 0,68 fps encoding
> .....) 20% can be a few hours difference, but don't know if without
> that option it will do the same good job

It only writes out the bitstream, which has completely no purpose
in this use case.
I implemented that.


More information about the MPlayer-users mailing list