[Ffmpeg-devel] svn 6835 crash in mpeg12.c

Aaron Williams aaronw
Sun Oct 29 21:46:23 CET 2006


Michael Niedermayer wrote:
> Hi
> On Sun, Oct 29, 2006 at 10:49:05PM +0300, Ismail Donmez wrote:
>> Hi,
>> 29 Eki 2006 Paz 20:43 tarihinde, Aaron Williams ??unlar?? yazm????t??: 
>> [...]
>>> If I compile mpeg12.c with -O2 instead of -O3 then it seems to work
>>> fine.  I am compiling with GCC 4.1.0 (SUSE Linux) for an Athlon64
>>> (-farch=athlon64).
>> You should file a gcc bug at gzz bugzilla.
> but before you do that try make distclean and check with valgrid
> that the -O2 code really is ok
> IMO we shouldnt randomly dump all not fully understood bugs into gcc bugzilla
> [...]
I am seeing a ton of valgrind errors with this release:

==24793== Source and destination overlap in memcpy(0x5FB48E0, 0x5FB48E0, 88)
==24793==    at 0x4A220D3: memcpy (mc_replace_strmem.c:116)
==24793==    by 0x4724D5: avcodec_default_release_buffer (utils.c:346)
==24793==    by 0x47F303: MPV_common_end (mpegvideo.c:464)
==24793==    by 0x5525A9: mpeg_decode_end (mpeg12.c:3207)
==24793==    by 0x473BA4: avcodec_close (utils.c:961)
==24793==    by 0x4262F0: av_find_stream_info (utils.c:1931)
==24793==    by 0x419656: opt_input_file (ffmpeg.c:2616)
==24793==    by 0x4214BB: parse_options (cmdutils.c:101)
==24793==    by 0x41F35F: main (ffmpeg.c:3958)
==24793== Conditional jump or move depends on uninitialised value(s)
==24793==    at 0x524DCA0: x264_slicetype_mb_cost (in
==24793==    by 0x524E5C1: x264_slicetype_frame_cost (in
==24793==    by 0x524F1E4: x264_rc_analyse_slice (in
==24793==    by 0x5270872: (within /usr/local/lib/libx264.so.54)
==24793==    by 0x5271A83: x264_ratecontrol_start (in
==24793==    by 0x528AD0E: x264_encoder_encode (in
==24793==    by 0x7A0167: X264_frame (x264.c:89)
==24793==    by 0x473B1E: avcodec_encode_video (utils.c:864)
==24793==    by 0x41E62F: output_packet (ffmpeg.c:825)
==24793==    by 0x420FCD: main (ffmpeg.c:1940)

I am upgrading x264 to the latest SVN and compiling it with -g to
hopefully include more information.  I will also try upgrading to the
latest gcc 4.1 svn and see what happens.


More information about the ffmpeg-devel mailing list