[MPlayer-users] mencoder problems : lame and 'make test' results

[pl]

On Wed, Nov 14, 2001 at 04:01:34PM +0200, Arpi wrote:
[...]
> > AFAIK there is no newer version of lame available.
> i've seen 3.90 beta mentioned somewhere (maybe the cvs version?)
> 
> > FYI: the new machine uses 2.96-98, the old one
> > uses 2.96-81 from RedHat 7.1. Never had any
> > problems with these two regarding "mplayer".
> maybe yet another bug of that "perfect" "best" compiler...
[...]
> > Has anyone another suggestion on what's going wrong
> > here?
> bad compiler or bug in lame

I've run some tests for lame 3.89beta compilation on the configurations
I had available ( Mandrake 8.0 gcc-2.96-0.48mdk / Debian stable gcc
2.95.2-13 / Debian unstable gcc 2.95.4-8 / Debian unstable gcc 3.0.2-3)

The test is braindead:
  make distclean ; ./configure && make test

In lame makefile, may be read: 
  " Because of compiler options and effects of roundoff, the number of
  bytes which are different may not be zero, but should be at most 30."

Most 'make test' results were above 400 !

I've checked a bit further with the CFLAGS. My conclusions after a 10's
of tests is that lame should not be compiled with default optimization flags
if quality (functionnality) matters.
  export CFLAGS=-O1
  ./configure
  make

Then I got a value of 20 with any of those compilers (YMMV).

Today's lame CVS and gcc 2.95.4 gives:
  -O0 => difference of 0 !
  -O1 => difference of 5 !
  default flags (-O3) => 402

PS: FWIW, the gcc296 I found does not even^W run the test with standard
flags, other versions may probably do it...
--
Best regards,
  pl