- Mb's were not skipped at the end of a GOB
- One value too much was used in iDCT
- Chroma_qscale was not updated for mquant
patch by ("Chris Flerackers" <cflerackers at androme dot be>)
Originally committed as revision 3332 to svn://svn.ffmpeg.org/ffmpeg/trunk
no malloc(64) memcpy free stuff
no filter1 A->B then filter2 A->B (yes not B->A)
no incorrect rouding after the 1d filter
Originally committed as revision 3177 to svn://svn.ffmpeg.org/ffmpeg/trunk