There were some comments earlier about the (lack of) good 
optimization in the gcc back-end for NetBSD/macppc.  The following 
would seem to be relevant for those who care:

>Date: Mon, 2 Jul 2001 10:35:51 -0700
>To: scitech@lists.apple.com
>From: Kenneth Prager <prager@ieee.org>
>Subject: Re: C compilers
>
>At 11:23 PM -0700 6/29/01, scitech-request@lists.apple.com wrote:
> >Message: 2
> >Date: 29 Jun 2001 16:10:41 -0400
> >Subject: C compilers
> >From: "Tom Keyes" <keyes@chem.bu.edu>
> >To: scitech@lists.apple.com
> >
> >Absoft pro fortran for OSX does not include a C compiler, presumably
> >because OSX already has gcc. One of my students did a very preliminary
> >benchmark comparing the same code compiled with gcc-OSX vs absoft C-OS9 on
> >a G4/450 and found absoft-OS9 about 1.5X faster. If that stands up it's a
> >problem. Has anyone else compared those C compilers? Could gcc really give
> >substantially poorer performance tha absoft on G4?
>
>Awhile back I was benchmarking FFTW on my 500 MHz DP-G4, running OS
>X.  Initially, the only compilers flags (CFLAGS) I used were...
>
>    gcc -O3
>
>I found that FFTW ran much slower than I expected.  Steven Johnson,
>one of the co-authors, suggested using the following CFLAGS for gcc,
>under OS X...
>
>    gcc -O3 -fomit-frame-pointer -fno-schedule-insns 
>-fschedule-insns2 -mcpu=750
>
>This resulted in about a 4x performance improvement.
>
>Ken Prager


Cross my heart, strike me dead, stick a lobster on my head.
John Crichton -- Farscape, 6/15/01
h.b.hotz@jpl.nasa.gov, or hbhotz@oxy.edu