CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   OpenFOAM Installation (http://www.cfd-online.com/Forums/openfoam-installation/)
-   -   Is march%3dopteron switch necessary for intel (http://www.cfd-online.com/Forums/openfoam-installation/57288-march-3dopteron-switch-necessary-intel.html)

mirko May 1, 2008 14:39

Hi, I am compiling the -dev
 
Hi,

I am compiling the -dev branch of OF (as of today, it's been a week, but I am pretty sure that I am getting somewhere -- famous last words).

I am doing this on a 64-bit RHEL with two dual core intel processors. Does the -march=opteron flag that I see invoked hurt, help, or not matter?

Thank you,

Mirko

dkingsley November 10, 2008 13:06

OpenFOAM-1.5.x also has the -m
 
OpenFOAM-1.5.x also has the -march=opteron as the default setting and I have been researching the effects of this setting on Xeon emt64 processors for the last few days.

Today I came acrross this article: http://software.intel.com/en-us/articles/effects-of-architecture-selection-switc hes

which seems to indicate that Intel users are lossing ~17% with the default setting.

The next time I merge the 1.5.x sources I am going to try -march=nocona.

Has anyone used the switch -mfpmath=sse with OpenFOAM?

mattijs November 10, 2008 14:13

We've tried various and differ
 
We've tried various and differences are negligible. The only one that really matters is -ffast-math which we don't put on by default since it disables floating point error trapping (sigfpe).

There is no option on gcc (as far as I know) to enable floating point reordering but still enable floating point error trapping.

If you have different findings let us know.


All times are GMT -4. The time now is 09:33.