Proebsting's Law

[Steve Mynott]

On Wed, Dec 14, 2005 at 12:00:23PM +0000, Paul Makepeace typed:

> Perl people cry "benchmark!" (or pedantically, "profile!") as soon as
> optimizations are mentioned but has anyone here actually done it with
> gcc??
> 
> Pick some code, gcc -O3 and gcc it and run them.

Whenever I have experimented with exotic flags (-O3, -march etc) in gcc
I have seen little difference in the (probably crude and misleading)
benchmarks and applications I have run.  

Neither have I seen much speed up between major releases.

However, I have noticed the alleged "optimisations" in every major
release of gcc have led to very much increased compile times.

The Tiny C Compiler (tcc) is an order of magnitude faster at compiling than
gcc.  Sure it produces poor code compared to gcc (which itself is supposed to
be poorer than payware).  But who cares since you can compile ten times as
much freeware in an evening than before??  

(OK it doesn't compile Perl yet but Lui worked fine...)

http://fabrice.bellard.free.fr/tcc/

Nothing real is CPU bound to the degree of a Mandelbrot set and I bet that was
one of the gcc target benchmarks and that speedup is as good as it gets!

-- 
Steve Mynott <steve at tightrope.demon.co.uk>