>>Not quite the end. We still don't have consensus on whether we should
>>be compiling everything with -m486, and no one has given any facts on
>>what this does to Pentium performance.
>
>	ftp://ftp.scl.ameslab.gov/pub/HINT/serial/

I've done some initial runs of this on a P90 machine overclocked to 100Mhz.
Thats a 66Mhz external clock, 32 meg of 60ns ram, 256K pipeline cache.  Using
netbsd 1.1, gcc 2.7.2, on a quiet (not silent) machine.

The initial findings are that -m486 doesn't alter results significantly, on
this benchmark, when used alone.  However, in combination with -ffast-math,
-m486 produces better results than -ffast-math alone.  Furthermore,
-malign-double causes a performance *hit*, but this is mitigated somewhat
by using -m486.  The combination of -m486 -ffast-math produced the best times.

I'm curious about the effect of -malign-double, performance reduction on a
pentium is contrary to the gcc info docs.  

-john