[MITgcm-support] Porting MITgcm to Solaris x64

[Constantinos Evangelinos]

On Thu 31 May 2007 11:56, Martin Losch wrote:

> Hi Bill,
>
> On 31 May 2007, at 17:50, Patrick Heimbach wrote:
> > There are differences between my output.txt and results/output.txt.
> > Some of these appear to be formatting differences between versions
> > checkpoint57c_pre and checkpoint58.
>
> You cannot expect exact correspondence if you run on a platform with
> a compiler that is different from the one that produced the
> output.txt files. See http://mitgcm.org/testing.html for a few examples.
>
> > I am getting a new message to the screen:
> >
> > STOP: NORMAL END
> >  Note: IEEE floating-point exception flags raised:
> >     Inexact;  Invalid Operation;
>
> I get this message on Solaris as well, and tried to trace it down. I
> failed because the Solaris compilers allow floating point traps to
> stop the code for debugging, they choke on a few tricks that the
> MITgcm does, so I never got anywhere. These messages may be related
> to some things that happen in the overlaps and do not have any effect
> on the solution.

Regarding the Inexact exception that is expected to be raised in anything 
other than floating point operations that have a result that can be exactly 
represented in the floating point format used without any rounding error 
(like 1.0+1.0). 

The Invalid Operations do come from the overlaps in my experience and involve 
0./0. type of operations during initialization (in the case of ECCO 
applications). There are also divisions by zero in some cases. At some point 
we should clean them up.

One can control floating point exceptions etc. for MITgcm using the -gsl flag 
to genmake2 and the GSL_IEEE_MODE environment variable - for examples see: 
http://www.gnu.org/software/gsl/manual/html_node/Setting-up-your-IEEE-environment.html

Constantinos
-- 
Dr. Constantinos Evangelinos
Department of Earth, Atmospheric and Planetary Sciences
Massachusetts Institute of Technology