calling MPI of Fortran from python - parallel-processing

Question

I try to calling MPI of Fortran from python.
in helloworld.f90, I write:
subroutine sayhello(comm)
use mpi
!include 'mpif.h'
implicit none
integer :: comm, rank, size, ierr
call MPI_Comm_size(comm, size, ierr)
call MPI_Comm_rank(comm, rank, ierr)
print *, ’Hello, World! I am process ’,rank,’ of ’,size,’.’
end subroutine sayhello
And another one called hello.py:
from mpi4py import MPI
import helloworld
fcomm = MPI.COMM_WORLD.py2f()
helloworld.sayhello(fcomm)
However, in the ubuntu 18.04, I use Python 3.7.4, I can not use the command to create file.so:
f2py -m helloworld -h helloworld.pyf helloworld.f90
f2py -c helloworld.pyf hello.py
When I run command:
mpirun.openmpi was unable to find the specified executable file, and
therefore
did not launch the job. This error was first reported for process
rank 0; it may have occurred for other processes as well.
Can you help me to fix this error or provide other command to run
My computer can openMP of Fortran from Python. Also, I can run MPI in Fortran . However, I can not call MPI of Fortran from Python.
Any help will be appreciated

Answer 1

I just want to make some comments about this issue. I create the Python module in Ubuntu 18.04. Actually, it is a WSL (Windows sub system). I run with openMPI 1.10.7 (is too old), but it is the version that I have. My tests were made using numpy 1.16.5 and Python 2.7.15.
1) I have created a Python module using your source codes through the following command from mpi4py doc page
f2py --f90exec=mpif90 -c helloworld.f90 -m helloworld
as said by Gilles you need to link the mpi, but you can use the flag --f90exec as shown above. This flag tells to f2py which F90 compiler you want to use. The f2pydoc has a complete list of f2py flags that will help you to create Python modules from Fortran codes. After the Python module generation you can run your script.
Another way to do it is like you did. However, I think that your second command is wrong. You do not use the flag --f90exec in the second command. Moreover, you have used the signature files with the name of your Python script in order to create the module. You should replace the Python script by the Fortran filename.
2) I tested what you have done. Then, when I tried to import the module in the Python interpreter a n exceptionwas raised. To sum up, in order to create a Python module using .pyf files type
f2py -m helloworld -h helloworld.pyf helloworld.f90
and then
f2py --f90exec=mpif90 -c helloworld.pyf helloworld.f90
3) King said that you do not need mpi4py. However, I tested using without mpi4py and did not work, even doing all the MPI initialization process inside a FORTRAN subroutine. Below, one can see the codes that I used to make this test.
helloworld.f90
subroutine sayhello
use mpi
implicit none
! include 'mpif.h'
integer :: comm, rank, size, ierr, namelength
character(len=15) :: processorname
call MPI_INIT(ierr)
call MPI_Comm_size(MPI_COMM_WORLD, size, ierr)
call MPI_Comm_rank(MPI_COMM_WORLD, rank, ierr)
call MPI_GET_PROCESSOR_NAME(processorName, namelength, ierr)
print *, 'Hello, World! I am process ',rank,' of ',size,'.'
end subroutine sayhello
hello.py
from mpi4py import MPI
import helloworld
helloworld.sayhello()
4) If you want to create a module for Python 3.x scripts you can use similar commands, you just have to substitute f2py by f2py3 or python3 -m numpy.f2py in the commands above.