[rescue] Re: Being jobless

[Charles Shannon Hendrix]

On Tue, Jul 29, 2003 at 07:43:11PM -0400, vance at neurotica.com wrote:
> On Tue, 29 Jul 2003, Charles Shannon Hendrix wrote:
> 
> > > > The binaries on my AMD Athlon system are now faster and smaller, so
> > > > I'm having a hard time complaining right now.
> > >
> > > That -is- my complaint.  The ONLY platform on which GCC has seen
> > > significant improvement is x86.
> >
> > Oh, I agree, but I thought the original sentiment was that it wasn't
> > optimized for *any* platform.
> 
> It's always been fine for 68K, but try using it for MIPS, or POWER, or
> Alpha.

I have... :(

Alpha was pretty bad.

But then, in 1996 we had so many troubles with DEC's C compiler,
we moved to gcc to *improve* things.  The biggest problem was some
outstanding math bugs, and a few issues with compatiblity.  DEC never
would acknowledge the problem, even though it was easy to repeat and
demonstrate.

In 1995 I used gcc on SunOS to generate code for 68040's running VxWorks
on VMEbus.  The compiler sometimes would spit out 32-bit writes to
memory on 16-bit integers.  This wreaked havoc with code that talked to
hardware.  Wind River sent us sources to fix the problem.  I'd not have
figured it out had I not had gcc spit out assembly so I could see what
was going on in detail.

This points out one of the reasons why 68K and x86 gcc code is good: the
vendors actively worked with the sources and issued fixes when problems
came up.

This has never happened consistently for the other architectures.

-- 
UNIX/Perl/C/Pizza____________________s h a n n o n at wido !SPAM maker.com