[XviD-devel] If I'm trying to compare differences between versions of XviD binaries...

[Edouard Gomez]

suxen_drol (suxen_drol at hotmail.com) wrote:
> On Fri Jun  6 06:38:15 2003 jmillenbach at hotmail.com (Joe Millenbach) wrote:
> > Can I just do a binary compare?  In other words, should I get exactly the 
> > same file out of XviD every time for a given input file and settings?  I'm 
> > not changing any source files, only adjusting optimization settings before 
> > building...
> 
> it'll be mostly identical, although this depends highly on the compiler/linker.

encoder output must be identical whatever compiler you use. Even
compiler options must not change the result.

For the xvid 0.9.1 release, i did lot and lot of testing on very exotic
platforms, and my tests were quite simple:
 0/ Choose a platform
 1/ make XviD compile on the platform.
 2/ xvid_encraw a choosen image sequence (first cactus and then some
 larger/longer sequence)
 3/ obtain the same encoded stream as the "reference" ia32 results
 4/ if 3/ fails find bugs goto 1/
 5/ if any access to other platform is available goto 0/

This simple  steps worked  fine for the  alpha platform, Strong  arm and
Ultra Sparc.  All versions  were buggy first  (mostly due  to endianness
problems   or  pointer<>int  conversion   performed  in   the  bitstream
functions)

Of course i was only comparing plain C versions. Assembly optimized code
paths introduce small rouding errors at various stages and thus do not
allow direct comparison (nor any other simple comparison, other than
result watching).

@pete: i do not think he was talking about xvid binaries comparison per
se ;-) It has no sense to compare code (except when some company try to
steal your hard work ;-)

-- 
Edouard Gomez