[MITgcm-support] Package global_ocean.cs32x15

[Ed Hill]

On Thu, 2005-04-28 at 16:52 -0230, Entcho Demirov wrote:
> Hi Steph, Ed et al.
> 
> I try to activate global_ocean.cs32x15 package on an alpha_linux_g77
> computer. I use the standard packages.conf file in the code
> subdirectory to generate Makefile and to compile the code. I have
> tried to work with the both OPTFILES g77 and g77+mpi and got one
> the same problem.
> 
> The run crashes already at the initialization step giving an error
> "Floating Exception". Since this kind of errors may occur everywhere,
> I have tried to trace where the model stops. I have found that the
> problem is related with the initialization of the the cubed-sphere
> grid in
> 
> file exch2_send_rs2.F
> 
> Subroutine EXCH2_SEND_RS2,
> 
>   line 146-147
> 
>            val1=sa1*array1(isl,jsl,ktl)
>       &       +sa2*array2(isl,jsl,ktl)
> 
> It looks like that in all of the calls to this subroutine,
> array1 and array2 are not initialized. Correspondingly the
> value of val1 may vary from very small (O(1.e-100)) to very
> large values (O(1.e300)). However the problem with the run
> occurs only when the value of one of these arrays is NaN,
> which produces the floating exception.
> 
> Do you have an idea how I may work around this problem?


Hi Entcho,

Yes, I do know one way to work around this problem and it might help
you.  Do you have access to a commercial (that is, an official "HP" or
"Compaq" or "DEC") compiler for the Alpha architecture?

Most of the commercial compilers for the alpha architecture have an
option that allows floating point exceptions to result in warnings
rather than errors that cause the program execution to halt.  And,
unfortunately, I don't think there is any way to make the Gnu g77
compiler behave that way -- it produces code that always halts when
floating point exceptions occur.

For instance, the "osf1_alpha_f77" optfile has:

  FFLAGS='-fpe2 [...]'

where the "-fpe2" flag tells the Fortran compiler to produce code that
will ignore floating point exceptions (actually, it completely ignores
all floating point exceptions after the first or second have happened).

If you don't have access to one of the commercial alpha compilers, then
it may be possible to re-write parts of MITgcm so that these floating
point exceptions don't happen for you.  But that may (or may not?)
entail a lot of work.

Ed

-- 
Edward H. Hill III, PhD
office:  MIT Dept. of EAPS;  Rm 54-1424;  77 Massachusetts Ave.
             Cambridge, MA 02139-4307
emails:  eh3 at mit.edu                ed at eh3.com
URLs:    http://web.mit.edu/eh3/    http://eh3.com/
phone:   617-253-0098
fax:     617-253-4464