[MITgcm-devel] SIZE.h matters on Columbia for CS510?

[Dimitris Menemenlis]

removing -O3, which is what Hong tried first, is equivalent to -O2,  
and it also failed.

so we have code that works with -O0, fails with -O2 and -O3, except  
for a specific
number of processors (270 CPUs), but I think that's just random.

This is on Columbia (Altix) and we are using the following optfile:
MITgcm_contrib/high_res_cube/code-mods/linux_ia64_ifort+mpi_altix_nas
which is a modified version of
MITgcm/tools/build_options/linux_ia64_ifort+mpi_altix_nas

Compiler is intel-comp.10.1.013

Hong just verified that all the sea ice output is OK at second time  
step so that
probably eliminates Martin's hypothesis.  Only Eta, seems to have  
problems at
beginning of second time step.

The next thing that we are trying is to use intel-comp.10.1.015

Dimitris Menemenlis
DMenemenlis at gmail.com

On Jun 27, 2008, at 12:56 PM, Jean-Michel Campin wrote:

> Hi Dimitris,
>
> Do you need to completly turn off the optimisation (-O0) or
> may be just reducing it  (-O3 -> -O2) would work ?
> On which machine & optfile are you working ?
> Cheers,
> Jean-Michel