[MITgcm-devel] compiler optimization on SunOS

[Martin Losch]

Sorry, some additional information. We use non-hydrostatic code.

Redoing the experiment on 6 CPUs with no tiling in x-direction is OK,  
so maybe that's what's missing in the verification experiments?

Martin

On 16 Feb 2007, at 12:35, Martin Losch wrote:

> Hi,
>
> I have a problem on our SunFire:
> SunOS model 5.9 Generic_112233-11 sun4u sparc SUNW,Sun-Fire-15000
> with
> f90: Sun Fortran 95 8.0 2004/07/15
>
> The testreport runs every week without any serious problem, see  
> http://mitgcm.org/testing.html
> For the tests I use MPI and -O4 as in the build options file  
> sunos_sun4u_f77+mpi_sunfire
>
> Now we have a configuration with 24CPUs, cartesian grid (no cubed  
> sphere (o:) which crashes for any optimization level >2 (-O3 is the  
> "recommended" optimization level according to the man pages).
>
> I have tried the list of NOOPTFILES from some very old genmake file,
> NOOPTFILES = barrier.F different_multiple.F external_fields_load.F  
> ini_vertical_grid.F ini_spherical_polar_grid.F ini_cori.F  
> mon_printstats_rl.F mon_printstats_rs.F
> but that doesn't help.
>
> Any further suggestions to add to the list of NOOPTFILES?
>
> Any idea, what might be going wrong that's not caught by the  
> testreport tests? I am attaching a STDOUT.0000 file.
>
> Puzzled as usual
>
> Martin
>
> <STDOUT.0000.gz>
>
>
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel