[FFmpeg-devel] dealing with tables in DV codec

Roman V. Shaposhnik rvs
Thu Sep 11 04:52:42 CEST 2008


On Thu, 2008-09-11 at 03:39 +0100, M?ns Rullg?rd wrote:
> >> By looking at the difference between two frames? Or is such case
> >> also handled in ffmpeg?
> >
> > That goes way over my head. DV is an intra-frame codec, so what
> > exactly do you mean by "'caching' of motionless blocks"?
> 
> I'm guessing here, but maybe it could, at least in some cases, be
> faster to compare a coded block with the corresponding coded block
> from the previous frame, and if they are the same, simply copy the
> pixels instead of decoding the block again.

Could that really be all that much faster. Especially for HD sizes
where each frame would occupy 4Mb of L1 cache? It is an interesting
though, though. Of course, for DV there could be an extra complication
of super blocks. And entire super block would have to be sort
of motionless and given that 5 macroblocks are quite shuffled that
might not be as frequent as one would like.

I guess, one way to know for sure is to see how much time a
straight-through iDCT would take on pre-reshuffled and decoded
macroblocks.

Thanks,
Roman.





More information about the ffmpeg-devel mailing list