[Ffmpeg-cvslog] r5422 - trunk/libavcodec/ratecontrol.c

[Corey Hickey]

Rich Felker wrote:
> On Mon, May 29, 2006 at 01:49:04PM -0700, Corey Hickey wrote:
>> michael wrote:
>>> Author: michael
>>> Date: Mon May 29 18:50:37 2006
>>> New Revision: 5422
>>>
>>> Modified:
>>>    trunk/libavcodec/ratecontrol.c
>>>
>>> Log:
>>> fabs is better then abs for floats ...
>>>
>>>
>>> Modified: trunk/libavcodec/ratecontrol.c
>>> ==============================================================================
>>> --- trunk/libavcodec/ratecontrol.c	(original)
>>> +++ trunk/libavcodec/ratecontrol.c	Mon May 29 18:50:37 2006
>>> @@ -911,7 +911,7 @@
>>>      av_free(qscale);
>>>      av_free(blured_qscale);
>>>  
>>> -    if(abs(expected_bits/all_available_bits - 1.0) > 0.01 ){
>>> +    if(fabs(expected_bits/all_available_bits - 1.0) > 0.01 ){
>>>          av_log(s->avctx, AV_LOG_ERROR, "Error: 2pass curve failed to converge\n");
>>>          return -1;
>>>      }
>> This is resulting in mencoder failing the second pass for me. Prior to
>> this revision, the same command works fine.
> 
> Looks like the threshold is wrong, and it just happened that the test
> was a virtual no-op before due to truncation.

Yes, something like that. I guess it should stop being lazy and post 
full steps to reproduce. I thought it was happening with every file, but 
it seems to only happen with some.

1. Compile mplayer CVS snapshot with SVN libavcodec.
2. $ wget http://fatooh.org/files/tmp/small.mpg
3. $ for i in 1 2 ; do mencoder small.mpg -nosound -ovc lavc -lavcopts 
vpass=$i:vmax_b_frames=2 -o /dev/null ; done

Also, if it matters, this is on amd64.

-Corey