Hi,
I am having a problem with OpenMPI and NetCDF4 Fortran 90 parallel IO.
I'm fairly certain the problem is with NetCDF and not HDF5 because 'make
check' for HDF5 tests the C AND Fortran 90 parallel IO and all tests
pass. I have traced this problem extensively in TotalView and it seems
to stem from the fact that MPI_COMM_WORLD is zero in the Fortran 90
interface for OpenMPI. In contrast, MPICH2 Fortran 90 interface sets
MPI_COMM_WORLD to 1140850688.
It errors on the first NetCDF function call, 'nf90_create_par'. The
error I get with OpenMPI is:
bash-3.00$ mpirun -np 2 nf90_test
[daggoo:12142] *** An error occurred in MPI_Comm_dup
[daggoo:12142] *** on communicator MPI_COMM_WORLD
[daggoo:12142] *** MPI_ERR_COMM: invalid communicator
[daggoo:12142] *** MPI_ERRORS_ARE_FATAL (your MPI job will now abort)
*** Writing output file : nf90_test.nc
---------------------------------------------------------------------
mpirun has exited due to process rank 0 with PID 12142 on
node daggoo exiting without calling "finalize". This may
have caused other processes in the application to be
terminated by signals sent by mpirun (as reported here).
---------------------------------------------------------------------
[daggoo:12141] 1 more process has sent help message
help-mpi-errors.txt / mpi_errors_are_fatal
[daggoo:12141] Set MCA parameter "orte_base_help_aggregate" to 0 to
see all help / error messages
bash-3.00$
If I compile and run the exact same program with MPICH2 I completes
successfully. There are several more NetCDF calls later in the program.
Of course, when using MPICH2 I use the HDF5 and NetCDF4 libraries
compiled with MPICH2.
Is there a workaround?
OpenMPI provides C functions to convert between C and Fortran values for
MPI_COMM_*:
MPI_Comm MPI_Comm_f2c(MPI_Fint comm)
&
MPI_Fint MPI_Comm_c2f(MPI_Comm comm)
Reference:
http://www.mpi-forum.org/docs/mpi21-report-bw/node355.htm#Node355
I have attached a gzipped tar of the two source files and the Makefile.
Thanks,
David Robertson
Attachment:
io_test.tar.gz
Description: GNU Zip compressed data
netcdfgroup archives: