[FFmpeg-devel] FFmpeg for Mac

[Vitor Sessak]

Neal Krawetz wrote:
> Hi folks,
> 
> I managed to port FFMpeg to Macs as a Universal Binary (runs on both ppc
> and x86 architectures).  There were a couple of changes that I needed
> to make in order for this work.
> 
> I've tested this on the x86 architecture.  It compiles for a PPC, but I
> have not tested it yet.  I suspect that it will fail because the ENDIAN
> is hard-coded in config.h.
> 
> Anyway, here is my notes of things I needed to change for porting to the Mac:
> 
> * Install ffmpeg.
>   Unforutnately, ffmpeg does not support universal binaries and configure
>   does not set all of the configuration variables correctly.  So we will
>   force it.
>   Edit libavformat/framehook.c and comment out the header check:
>         // #if HAVE_DLFCN_H
>         #include <dlfcn.h>
>         // #endif
>   This header DOES exist when Xcode for MacOSX 10.5 is installed.

Interesting. Is the configure check for this header failing (note that 
our configure script is hand-made, not autotools-generated mess, so it 
should be very easy to debug)?

>   Edit libavformat/rtpproto.c and comment out the header check:
>         // #if HAVE_SYS_SELECT_H
>         #include <sys/select.h>
>         // #endif
>   Edit ffserver.c and comment out the header check:
>         // #if HAVE_DLFCN_H
>         #include <dlfcn.h>
>         // #endif

Same thing.

>   Now compile:
>     CFLAGS='-arch ppc -arch i386' ./configure --arch=x86
>     CFLAGS='-arch ppc -arch i386' ./configure --arch=x86
>     # Yes: Run configure twice! There is a permissions problem for config.h.
>     make
>     sudo make install
>   This will install into /usr/local/.
> 
> 
> As far as the endian issue goes, have you considered using a global
> variable that sets the endian?  This would simplify portability.
> Here's what my code (Piana) does since I have this same issue:

This is not possible for speed reasons. FFmpeg is full of very speed 
critical code like

for (i=0; i < frame_size; i++)
     out[i] = table[bytestream_read_le32(&in)];

where the implementation of bytestream_read_le32 is endianness dependent 
(and we cannot afford an extra if()). There is much more code like this 
(all bytestream readers, bitstream readers, scalers, etc).

-Vitor