[Ffmpeg-cvslog] CVS: ffmpeg/libavcodec vp3.c,1.64,1.65

Michael Niedermayer michaelni
Sat May 21 18:40:53 CEST 2005


On Saturday 21 May 2005 18:03, Mike Melanson wrote:
> Michael Niedermayer wrote:
> > i can, and the md5 of the output changes which is what matters, the md5
> > must not change due to optimizations unless its fully understood why
> > if this rule is ignored you quickly end up with a deeply broken mess
> > which is very hard to debug and fix
> 	This reminds me: When you deployed the loop filter, how did you even
> know that the output was correct?

i dont, it probably is not 100% correct, but i do know that your optimized 
code is worse it is vissible, maybe run a sharpen filter over the image if 
you dont see it

> >>but performance actually seems
> >>to go down with the optimizations enabled...
> >
> > you mean with this one too? the unconditional change to decoding all 3
> > planes at once instead of sequantially even if the user app doesnt
> > support slices did already reduce the speed ...
> 	Can I use the standard FFmpeg/MPlayer excuse and blame it on the compiler?

certainly, as long as the in practice fastest variant is used


More information about the ffmpeg-cvslog mailing list