[FFmpeg-devel] build failure on osx 10.5.8 ppc

[Pavel Koshevoy]

On 10/7/12 5:58 PM, Michael Niedermayer wrote:
> On Sun, Oct 07, 2012 at 05:44:47PM -0600, Pavel Koshevoy wrote:
>> On 10/7/12 4:06 PM, Michael Niedermayer wrote:
>>> On Sun, Oct 07, 2012 at 03:11:05PM -0600, Pavel Koshevoy wrote:
>> <snip>
>>> I added a NULL pointer de-reference and it didn't crash, so that
>>> tells me I am not exercising that code path.
>>>
>>> Any other suggestions, or should we give up on the benchmark?  I
>>> need to go clean the house and then work on a patch for fontconfig
>>> :)
>>> compress a mp3 of your choice to .m4a with ffmpeg and then
>>> try to decompress it, it should excercise the code
>>> or take aac/al04_44.mp4 from the fate suite as a random samplle that
>>> should use the code
>> I rebuilt ffmpeg as follows:
>> /nfs/scratch/Developer/ffmpeg-git-src/configure
>> --prefix=/Developer/ppc --enable-swscale --enable-avfilter
>> --enable-libmp3lame --enable-libvorbis --enable-libopus
>> --enable-libtheora --enable-libschroedinger --enable-libopenjpeg
>> --enable-libmodplug --enable-libvpx --enable-libspeex
>> --enable-shared --enable-pthreads --enable-gpl --enable-version3
>> --enable-libopencore-amrnb --enable-libopencore-amrwb
>> --enable-postproc --enable-libx264 --enable-libxvid
>> --enable-runtime-cpudetect --extra-cflags=-I/opt/local/include
>> --extra-ldflags='-headerpad_max_install_names -L/opt/local/lib'
>> --disable-stripping --enable-debug=3 --disable-optimizations
>>
>> I loaded ffmpeg_g into gdb in emacs and set a breakpoint in
>> float_to_int16_stride_altivec.  I converted an mp3 to an m4a (had to
>> add -strict -2 to ffmpeg options, and it was really slow).  I then
>> converted the output m4a back to mp3 -- the breakpoint (and the NULL
>> pointer de-reference were never hit).  I tried with aac/al04_44.mp4
>> from fate suite and that didn't hit float_to_int16_stride_altivec
>> either.  Looking at the code, float_to_int16_stride_altivec is only
>> called from float_to_int16_interleave_altivec if number of channels
>> is greater than 2.
>>
>> I also tried converting the 5.1 ch WMV file to flac again, and that
>> didn't enter this function either.  ffprobe says that WMV file has
>> flt samples, so it would seem to be a perfect candidate for this
>> test.
> this is strange it hit the x86 function here
> have you checked that the function pointer is set at all ?

Nope.


> and you can just try make fate with the null deref in there to check
> if any test uses it

OK, I'll leave the NULL de-reference and run fate, but it's going to 
take a while since this is a pretty slow machine (10 years old), and I 
also disabled optimizations for the debug build.


>
>
>> BTW, Is there a way to turn off ANSI colors on ffmpegs
>> pretty-printed output?  It's really annoying in emacs gdb mode (it
>> doesn't interpret the color codes, just spams the buffer with them).
> export AV_LOG_FORCE_NOCOLOR=1

Good to know, thanks.

     Pavel