[Ffmpeg-devel] [PATCH] Chinese AVS video decoder
Sun Jul 2 13:26:07 CEST 2006
new patch versus current SVN attached, hopefully addressing
all outstanding points.
Von: Michael Niedermayer
> the intra prediction code, or more precissely the border loading stuff
> is a little messy, first things are stored in left/top_border* then they
> are copied to top/left and then they are finaly used, can that 2 level
> copy be avoided and pointers to left/top_border* be passed into the
> intra prediction functions?
avoiding the first copy could be done by deblocking the whole frame after
decoding, but I think the H.264 code was changed the other way round
for good reason.
For the second copying: The advantage is that the mess is concentrated
in load_intra*, while all the intra prediction functions are only a few
lines and very readable. I did actually try to make these decisions about
neighbouring pixels in the prediction functions themselves at one point,
but after a few functions I didn't like the look of it. Sure, there is
a price on performance, but I think that stuff is not the most critical.
> ive added the new files to svn, ill commit the rest of your patch as soon
> as you add a ff_ prefix (or av*_ for anything which should be vissible to
> the user application) to the new non static functions
Done, with one change: Making those put/avg_pixels* functions in
dsputil.c non-static and prefixed with ff_ seemed to turn into quite an
intrusive patch, instead I moved four stub functions (the *_mc00 stuff)
from cavsdsp.c to dsputil.c
> also if you want to maintain the AVS decoder and want a svn account, just
> send a password and username to diego (and read the development docs &
I would like to maintain it, even though my response times might not be the
"Feel free" ? 10 GB Mailbox, 100 FreeSMS/Monat ...
Jetzt GMX TopMail testen: http://www.gmx.net/de/go/topmail
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 5800 bytes
Desc: not available
More information about the ffmpeg-devel