Re: Building Octave with Intel C++/F9x

[Mumit Khan]

On Sun, 9 Jun 2002, Hendrik Naumann wrote:

> I would be interesed in some benchmarks comparing with gcc2.x and/or
> gcc3.x. Does this effort realy pay?

I haven't done any benchmarking, other than watching the runtimes of some
of my octave code. The various fortran routines, when built with Intel
Fortran compiler, shows reasonable speedups over g77; a lot of my floating
point intensive code shows anything from -5% to +25%, with the mean around
15%, speedup over gcc-3.1 (all under RedHat 7.2, gcc-3.1, PIII box). I
would however caution that none of these tests were done with benchmarking
in mind, and one should not draw any specific conclusions based on this
data. I have yet to check if the optimizer bugs that plagued 5.x releases
have been fixed or not. Until I can tweak our configuration scheme to use
Intel's compilers without hand tweaks, I doubt if I'll do any stress
testing.

Regards,
Mumit