ROMS compilation on Solaris Opteron

[aingunza]

Dear friends:

I would like to compile ROMS on a SUN computer with 2 opteron processors (32/64 bits) and Solaris 10.0 i386 Operating System. I have installed the fortran 90/95 compiler (from SUN Studio 11) but I have some errors during the compilation.

I would like to know if anybody has already compiled ROMS with such a computer or if anybody knows what is the better compiler for ROMS with Solaris 10.0 i386 release 5.8.

Thank so much for your answers and your tips.

Augusto Ingunza

[kate]

I don't have this system, but ROMS should be fairly portable and has been run on a variety of systems. Please tell us what version you are using and what errors you are getting. You can save the errors by redirecting standard error (make >& make.out in csh).

[thaned]

I don't have this system either. However, I had problem compiling on Sun earlier. My problem was cpp on my machine didn't have everything need. I found out the I needed to install a special cpp from the following link.

viewtopic.php?t=87&highlight=cpp

I am not sure, this may or may not solve your problem.

[mab]

I don't have this system, but ROMS should be fairly portable and has been run on a variety of systems. Please tell us what version you are using and what errors you are getting. You can save the errors by redirecting standard error (make >& make.out in csh).

We have been using a 12 processor/24 node cluster to run ROMS at
VIMS for about a year. The cluster is made up of 12 SUN v20z 2-node Opteron based computers. We use MPICH 1.2.6 and the SUN Studio 11
compilers. NETCDF is version 3.6.0. The OS is Solaris 10. We have
both GigE and Infiniband switches.

Courtney Harris' group is working with Rich Signell
on the Sediment code using the ADRIA02 model.

We compile consistently for MPI using
FFLAGS=-g -O3 -xterm=opteron -xarch=amd64

FFLAGS=-g -O5 -xterm=opteron -xarch=amd64 produces the same
output, but doesn't give much speedup and deviates from IEEE
standards. The -fast flag gives output that differs from the -O3 and the -O5 flags, as well as violating IEEE standards.
The times are about the same for -O3 and -O5, so we have stayed
with -O3. Using -xarch=amd64a doesn't give any advantages.

If we try to omit the -g flag that includes the symbol table, the
ROMS code fails in step2d.f with a floating point overflow. We haven't
tracked down the cause of this behavior.

The ROMS code has also been compiled and run with OpenMP
using shared memory on a SUN v40z with four processors. The
same FFLAGS were used. It will also run in serial mode when
compiled with the same flags.

Problems that occurred when the cluster was first installed were
traced to the compilation of MPICH and NETCDF, as well as the
patches on the OS.