MPI Fortran code error on a single PC

Question

Recently I was trying to compile and run my mpi code on a single machine (Ubuntu 12.04 - 64 bits core i7 2670 QM) I installed mpich2 version 1.2 using the following configuration:

./configure --prefix=/opt/mpich2 --enable-f77 --enable-fc --enable-cxx --with-device=ch3:sock --with-pm=mpd CC=icc CXX=icpc F77=ifort FC=ifort 2>&1 | tee configure.log

The installation was ok, and I got mpd working well, I tested mpd with the examples and all is perfect.

I compile my code using mpif77 because I don't know why when I compiled mpich2 mpif90 was not created. But even if with mpif77 I got the code compiled with no errors.

The flags I'm using to compile the code are:

For the compiler:

LN_FLAGS= -lm -larpack -lsparskit -lfftw3 -lrt -llapack -lblas

For MPI linker:

LN_FLAGS_MPI= $(LN_FLAGS) -I$(MPIHOME)/include -L$(MPIHOME) $(MPIHOME)/lib/libmpich.a -lfmpich -lopa -lmpe

So the problem is when I try to run the code on my machine:

First I invoke mpd as:

mpd &

and then run the code as:

mpirun -np 4 ./code_mpi

I tried a lot of variations as:

mpiexec -np 4 ./code_mpi
mpirun -n 2 ./code_mpi
mpiexec -n 2 ./code_mpi

And all results in the same error:

Fatal error in MPI_Comm_rank: Invalid communicator, error stack:
MPI_Comm_rank(106): MPI_Comm_rank(MPI_COMM_NULL, rank=0x14b46a0) failed
MPI_Comm_rank(64).: Null communicator
Fatal error in MPI_Comm_rank: Invalid communicator, error stack:
MPI_Comm_rank(106): MPI_Comm_rank(MPI_COMM_NULL, rank=0x14b46a0) failed
MPI_Comm_rank(64).: Null communicator
[cli_2]: aborting job:
Fatal error in MPI_Comm_rank: Invalid communicator, error stack:
MPI_Comm_rank(106): MPI_Comm_rank(MPI_COMM_NULL, rank=0x14b46a0) failed
MPI_Comm_rank(64).: Null communicator
[cli_1]: aborting job:
Fatal error in MPI_Comm_rank: Invalid communicator, error stack:
MPI_Comm_rank(106): MPI_Comm_rank(MPI_COMM_NULL, rank=0x14b46a0) failed
MPI_Comm_rank(64).: Null communicator
rank 2 in job 1  ubuntu_38132   caused collective abort of all ranks
  exit status of rank 2: killed by signal 9 
Fatal error in MPI_Comm_rank: Invalid communicator, error stack:
MPI_Comm_rank(106): MPI_Comm_rank(MPI_COMM_NULL, rank=0x14b46a0) failed
MPI_Comm_rank(64).: Null communicator
[cli_3]: aborting job:
Fatal error in MPI_Comm_rank: Invalid communicator, error stack:
MPI_Comm_rank(106): MPI_Comm_rank(MPI_COMM_NULL, rank=0x14b46a0) failed
MPI_Comm_rank(64).: Null communicator
Fatal error in MPI_Comm_rank: Invalid communicator, error stack:
MPI_Comm_rank(106): MPI_Comm_rank(MPI_COMM_NULL, rank=0x14b46a0) failed
MPI_Comm_rank(64).: Null communicator
[cli_0]: aborting job:
Fatal error in MPI_Comm_rank: Invalid communicator, error stack:
MPI_Comm_rank(106): MPI_Comm_rank(MPI_COMM_NULL, rank=0x14b46a0) failed
MPI_Comm_rank(64).: Null communicator
rank 1 in job 1  ubuntu_38132   caused collective abort of all ranks
  exit status of rank 1: return code 1

I spent almost 2 weeks trying to solve this problems because I really need to run this code in my personal computer to work at home. I appreciate all that can help me!

Here is how I initialize the MPI Library

subroutine init()
integer                      :: provided
call mpi_init(mpi_err)
call mpi_comm_rank(mpi_comm_world,rank,mpi_err)
call mpi_comm_size(mpi_comm_world,an_proc,mpi_err)
call MPI_BARRIER(MPI_COMM_WORLD,mpi_err)
end subroutine init

You shouldn't need to start mpd first; and can you compile/run a simple MPI "Hello world" successfully, eg slac.stanford.edu/comp/unix/farm/mpi.html ? — Jonathan Dursi
– Jonathan Dursi, Commented Oct 19, 2012 at 11:14
Show us this part of the code where you initialise the MPI library, e.g. the part that contains the calls to MPI_INIT and to MPI_COMM_RANK. — Hristo Iliev
– Hristo Iliev, Commented Oct 19, 2012 at 11:33

mgilson · Accepted Answer · 2012-10-20 01:47:07Z

4

The problem is your subroutine has no idea what mpi_comm_world is. This integer value is set in the mpif.h header (or the mpi module for f90). As your code is written, mpi_comm_world is assigned randomly by the compiler and has no association with the actual mpi_comm_world communicator handle provided by mpi.

Generally, it's best to use implicit none in your code which will warn you of these types of errors. Try the following:

subroutine init()
!use mpi   !This one is for f90
implicit none
include 'mpif.h'   !use this for f77
integer  :: provided,rank,an_proc,ierr
call mpi_init(ierr)
call mpi_comm_rank(mpi_comm_world,rank,ierr)
call mpi_comm_size(mpi_comm_world,an_proc,ierr)
call MPI_BARRIER(MPI_COMM_WORLD,ierr)
end subroutine init

answered Oct 20, 2012 at 1:47

mgilson

312k70 gold badges656 silver badges722 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Carlos Antonio Ribeiro Over a year ago

Really strange, because when I run on a cluster, I don't get this error. I made the changes and the error continues...

Collectives™ on Stack Overflow

MPI Fortran code error on a single PC

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related