some remarks about gcc4.0

[tbp <tbptbp at gmail dot com>]

First i must say that i got bit worried by
http://gcc.gnu.org/ml/gcc/2005-01/msg00268.html because while i've
seen remarkable progress being made lately to the generated code
quality on x86/SSE it's still quite brittle.

I've been implementing a coherent ray tracer, in parallel with Jacco
Biker. Mine was primarily developped with GCC (but also supports msvc)
and lags a bit behind, his with Intel's.
Today i've helped him to port it to gcc4.0 and in less than 2 hours
got it to match Intel's performance, admittedly with a few kludges.

You can see his Intel compiled demo here:
http://www.flipcode.com/cgi-bin/fcarticles.cgi?show=65091

This demonstrate that gcc can stand toe to toe with Intel in
applications making heavy use of floating points and vectors.
But right now it still requires voodoo magic, sacrificing a couple of
goats and a raindance before each compilation.

Please tell me that x86/SSE won't be left on the side of the road for
this release :)

	T. Berger-Perrin.

PS: many thanks for the gcc wizards and their prompt bug fixing.
PPS: msvc2k3 is typically more than 25% slower.