[Ffmpeg-devel] [PATCH] Chinese AVS video decoder
Sat Jul 1 13:05:44 CEST 2006
second try of a patch, following points addressed:
- reuse frame_rate_tab from mpeg12.c
- use ff_zigzag_direct as scan table
- simplified intra deblocking as Michael suggested
- create new file cavsdsp.c for DSP functions. In consequence, there is
also a header file for the prototypes and some put/avg_pixels8/16
functions in dsputil.c are made non-static. Is this acceptable?
- reduced size of constant tables
- output delayed frame when called with buffer size zero
- branchless mv_pred_direct, although it looks a bit obfuscated now
- moved all divisions in MV prediction to frame level
- fixed ffmpeg transcoding: disabled check_for_slice for now. It
is not needed in the sample clips available. Strangely, the
show_bits_long call in it returned wrong and different results
when running in ffmpeg rather than ffplay.
- Better handling of open GOPs by using ff_cavs_flush, this avoids
the artefacts after seek
- use get/release_buffer() from AVCodecContext
Points which are still open:
- AMD64 bug reported by Benjamin. All my 64bit knowledge comes
from Wikipedia :)
Is it correct to assume that I have to deal with LP64 data model?
So I only need to check for wrong assumptions about pointer size?
(I have not used any 'long')
- Doxygen comments.
Well, the code is written in C, ergo self-explaining.
But seriously, what are the requirements?
- ABSCMP macros suggested by Michael. I have not done any benchmarks
- Artefacts reported by Roberto and Matthieu
Roberto helped me a bit more off-list to circle in on it, but I am
still unable to reproduce. Maybe I fixed it by accident.
"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: 21910 bytes
Desc: not available
More information about the ffmpeg-devel