[XviD-devel] Great news: Bug fixed !

Edouard Gomez ed.gomez at free.fr
Fri May 23 01:35:09 CEST 2003


suxen_drol (suxen_drol at hotmail.com) wrote:
> > Pete any news for the 2pass Code,  does it work for you 100% of the time
> > ?
> no, i simply havent had time; i'll run some futher tests this weekend.

I found the bug, this was a stupid overflow in rc->target calculation.

Commited as CVS file version 1.1.2.9, it's an aggregated patch from my
tree. I just forgot to mention a fix to the min_size calculation that
was false due to a wrong test condition (pre_process0).

2003-05-22 GMT	Edouard Gomez <ed.gomez at free.fr>	patch-57

    Summary:
      Fixed an overflow bug in target filesize computation.
    Revision:
      xvidcore--devapi4--1.0--patch-57

    rc->target was an uint64_t data to avoid overflow when dealing
    with long movies and/or high bitrates. The problem is that its
    initialization was using int32 data, thus this was resulting
    in an overflow in its initial computation. Quite silly, but this
    bug drived me crazy during 4 hours...

    modified files:
     src/plugins/plugin_2pass2.c

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Now the cool results:
$MENCODER \
    -ovc xvid \
    -vop scale=640:346 -sws 9 \
    -xvidencopts
    debug=128:pass=2:bitrate=2048:quality=6:max_bframes=2:bquant_r...
    -oac mp3lame \
    -af volume=5:sc \
    -lameopts cbr:br=96:q=0:aq=0:mode=1 \
    Matrix.Reloaded-Trailer.mov

resulted in:
Video stream: 2049,577 kbit/s  (256197 bps)  size: 38771164 bytes
151,333 secs  3635 frames

And so was for 602, 512, 1024... Rate control just works !

Good isn't it ?

</me happy>

I think we can now start testing this tree more seriously.

-- 
Edouard Gomez-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://edu.bnhof.de/pipermail/xvid-devel/attachments/20030523/2866c95d/attachment.bin


More information about the XviD-devel mailing list