Optimisation and float problem

Optimisation and float problem

[lelievre tony]

I work with a finite element code in C++. This program works well on SGI (Octane R12000, Origin200 4 * R10000).

I have tried to compile it on a PC under Windows NT. The PC is a PC Compaq AP550 833 Mhz avec 1 Go de RAM. I use Cygwin and gcc-2.95. I also use gcc-2.95 on SGI.

The problem is the following :
- when I compile with the same option as on SGI (-O2), one of the calculation (inversion of a matrix by an iterative procedure : GMRES) leads to NaN result, whereas it works well on SGI.
- when I compile with no optimisation, it also works on PC, but the time of computation is doubled...

I would like to know which of the optimisation of -O2 I should turn off to make it work well. I have already tried -O2 -ffloart-store, with no success.

Or perhaps the problem may have an other solution...

Any clue ?

Thanks a lot,

Tony LELIEVRE.

Francemail 
Votre email gratuit.
http://www.francemail.com