[Ffmpeg-devel] assertion failure in epzs_motion_search_internal

[Michael Niedermayer]

Hi

On Thu, Dec 07, 2006 at 05:01:52PM -0800, Trent Piepho wrote:
> I'm getting an assertion failure:
> 
> mencoder: motion_est_template.c:1044: epzs_motion_search_internal: Assertion `(P[3][0]>>shift) <= xmax' failed.
> 
> This is when encoding some 1280x720 HD material with nearly latest svn,
> r7245.
> 
> Options used with mencoder are:
> 
> vcodec=mpeg4:qpel:v4mv:psnr:qprd:vbitrate=6950:vqmin=2:lmin=1:dc=8:trell:
> mbd=2:precmp=2:subcmp=2:cmp=2:dia=-4:predia=8:cbp:mv0:sc_factor=4:
> vmax_b_frames=2:vb_strategy=2:b_sensitivity=20:brd_scale=1:bidir_refine=1:autoaspect:vpass=1
> 
> I'm trying to find a sample that will trigger the error without 5 or 6
> hours of encoding first.
> 
> I had this problem before and was able to trace it up farther.
> 
> ff_estimate_motion_b() was setting P_TOPRIGHT[0] or P_TOP[0] to values that
> were out of range.  Around line 1517 in motion_est.c.  The values were not
> copletely invalid, but just invalid for that macroblock.  For example,
> something like 1200 for the P_TOP[0], which would be ok for some
> macroblocks in a 1280 pixel side source, just not for the one it
> was getting assigned too.

if "MV.x" is too large for the current macroblock then so it is for
the macroblock above and aboveright (P_TOP and P_TOPRIGHT)
so it _seems_ that an illegal motion vector has already been used in a
previous macroblock ...

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

The worst form of inequality is to try to make unequal things equal.
-- Aristotle