Bug#489732: ffmpeg-free: better gcc optimizing flags

[Reinhard Tartler]

tag 489732 pending
stop

A Mennucc <mennucc1 at debian.org> writes:

> following mplayer bug 489411, I was looking to see if FFmpeg may be
> optimized more; so I noted that the current gcc command line is quite
> messy; e.g. looking into
>  http://buildd.debian.org/fetch.cgi?pkg=ffmpeg-free;ver=0.svn20080206-8;arch=alpha;stamp=1212572490
> inside the line of  gcc for compiling motion_est.c ,
> I count two "-O2" and two "-O3"
>
> (does this make it "-O2.5" ? :-)

No, I think it takes the last switch, which is -O3

> Anyway, the same code, when compiled inside MPlayer, was compiled 
> with "-O4 -ffast-math -fomit-frame-pointer"
>
> Would it be possible to use the same optimization?

-O4 is pointless, it is the same as -O3. Which is generally frowned upon
in debian. Policy (more or less strongly) recommends to use -O2.

As for -ffast-math and -fomit-frame-pointer, I tend to trust upstreams
decision on optimization flags. I had a lenghty discussion with upstream
about that just today.

However I committed a fix to not errournously add both -O2 and -O3
compilation flags. the next upload at least should reenable MMX
optimized replacements, so mplayer should be a lot faster.

If someone has a non-mainstream arch to test, I'd love to hear build
reports from sparc, powerpc, alpha, mips, etc. for the current trunk for
ffmpeg-debian (that has these parts tweaked).


-- 
Gruesse/greetings,
Reinhard Tartler, KeyID 945348A4