[FFmpeg-cvslog] r11855 - in trunk: libavformat/nutenc.c tests/libav.regression.ref

[ods15 at ods15.dyndns.org]

On Mon, Feb 04, 2008 at 04:26:02PM -0500, Rich Felker wrote:
> On Mon, Feb 04, 2008 at 11:45:33AM +0100, ods15 wrote:
> > Author: ods15
> > Date: Mon Feb  4 11:45:32 2008
> > New Revision: 11855
> > 
> > Log:
> > Fix intented algo in syncpoint writing before keyframes in NUT muxer
> > 
> > 
> > Modified:
> >    trunk/libavformat/nutenc.c
> >    trunk/tests/libav.regression.ref
> > 
> > Modified: trunk/libavformat/nutenc.c
> > ==============================================================================
> > --- trunk/libavformat/nutenc.c	(original)
> > +++ trunk/libavformat/nutenc.c	Mon Feb  4 11:45:32 2008
> > @@ -462,7 +462,7 @@ static int write_packet(AVFormatContext 
> >      if(1LL<<(20+3*nut->header_count) <= url_ftell(bc))
> >          write_headers(nut, bc);
> >  
> > -    if(key_frame && !!(nus->last_flags & FLAG_KEY))
> > +    if(key_frame && !(nus->last_flags & FLAG_KEY))
> 
> This could explain the insane overhead lavf's nut muxer was
> generating... Have you tested to see if that's fixed now? :)

The insane overhead had more to do with r11854:
"nut->max_distance was not set at all anywhere in the NUT muxer"

max_distance was treated as zero. But this commit also caused huge 
overhead :)
Anyway, yeah, it's much lower now, I think as much as 0.5% . I'm not sure 
if libnut is a bit smarter in the framecodes it generates, worth 
checking...

- ods15