[FFmpeg-devel] [RFC] filter_mb_fast

[Michael Niedermayer]

On Thu, Dec 18, 2008 at 02:58:00AM -0800, Jason Garrett-Glaser wrote:
> Do we know under which cases filter_mb_fast is not bit exact?
> 
> I heard that it might not be in cases of duplicated reference frames
> or something of the sort, but I haven't gotten any sure information.

i can make a list of reference streams that fail with it enabled if you
like.
also CAVLC + 8x8dct also might be wrong, it at least needed a
fix in filter_mb() ...

and about reference, its not just duplicates in a list, its that lists
for slices can differ and that I/P/B slices can be mixed and that
matches across lists (in b slices) have to be considered.


> 
> One though of mine is that--AFAIK--filter_mb_fast uses the same logic
> as x264's deblocking, so we could easily activate it for now for at
> least x264 streams, knowing nothing else about why it isn't bitexact.

I am against a if(x264) filter_mb_fast() check. Correct is to check if
all the conditions are true when filter_mb_fast() is bitexact.


-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

There will always be a question for which you do not know the correct awnser.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20081218/62e9c300/attachment.pgp>