problem with parallel run GETM v2.0.0

32 views
Skip to first unread message

mahmood

unread,
Jan 31, 2012, 10:48:42 AM1/31/12
to getm-...@googlegroups.com
Dear all
I have a question and a problem on running GETM v2.0.0 in parallel mode
 first my problem:
I could run GETM v1.8 in parallel mode using "mpirun -np 2 ./bin/getm_prod_IFORT"
but now I am getting this error while trying to make my model:
((
halo_mpi.F90(18): error #7002: Error in opening the compiled module file.  Check INCLUDE paths.   [MPI]
   use mpi
-------^
halo_mpi.F90(40): error #6592: This symbol must be a defined parameter, an enumerator, or an argument of an inquiry function that evaluates to a compile-time constant.   [MPI_COMM_WORLD]
   integer, public                     :: comm_hd=MPI_COMM_WORLD
--------------------------------------------------^
.
.
.
.
compilation aborted for halo_mpi.F90 (code 1)

I have set these in my bashrc
PATH="/mirror/opt/mpich2/bin:${PATH}"; export PATH
PATH="/mirror/opt/mpich2/lib:${PATH}"; export PATH
PATH="/mirror/opt/mpich2/include:${PATH}"; export PATH
PATH="/mirror/opt/mpich2/sbin:${PATH}"; export PATH

and also in the make file: export GETM_PARALLEL=true
))
do I miss something?

second my question:
is it possible to have section with unequal wides? ( my domain is 583*123, is it possible to run it on 4 processors with 3 sections 134*123, and a 136*123 section?)

thank you

fatemeh chegini

unread,
Jan 31, 2012, 12:27:49 PM1/31/12
to GETM-users
Hi,

Have you done this?

export MPI=MPICH2
> second my question:
> is it possible to have section with unequal wides? ( my domain is 583*123,
> is it possible to run it on 4 processors with 3 sections 134*123, and a
> 136*123 section?)

I don't think that's possible since the halo zones would be messed up.


> thank you

You're Welcome,

Best,
Fateme

Knut

unread,
Jan 31, 2012, 12:54:56 PM1/31/12
to getm-...@googlegroups.com
Hi Mahmood,

> I have a question and a problem on running GETM v2.0.0 in parallel mode
> first my problem:
> I could run GETM v1.8 in parallel mode using "mpirun -np 2
> ./bin/getm_prod_IFORT"
> but now I am getting this error while trying to make my model:
> ((
> halo_mpi.F90(18): error #7002: Error in opening the compiled module
> file. Check INCLUDE paths. [MPI]
> use mpi
> -------^

getm-1.8 included header files instead of using module files.
I assume (but do not know) that in former times it was sufficient to
specify MPIINC. Now you definitely have to specify MPILIBDIR. ifort
needs these information to compile the mpi code.

Since you are using MPICH2 you only have to export MPI=MPICH2 (as
Fatemeh already suggested). Then instead of ifort the container mpif90
will compile GETM and inherently knows the necessary mpi information.

> second my question:
> is it possible to have section with unequal wides? ( my domain is
> 583*123, is it possible to run it on 4 processors with 3 sections
> 134*123, and a 136*123 section?)

GETM is designed to have the same size for all subdomains.

Knut

mahmood

unread,
Feb 19, 2012, 8:33:10 AM2/19/12
to GETM-users
thank you all
Reply all
Reply to author
Forward
0 new messages