example of FORTRAN code for MPI RMA operations (MPI_GET, MPI_put,...)

[Guillaume_De_Nayer]

Hi,

I'm reading a little bit about RMA operations and I'm trying to play with. However, I do not find some working examples of FORTRAN code, so that I can progress. If someone has some simple working examples in FORTRAN, please share :)

For the moment I only found the subroutines SUM and MAPVALS from mpi-forum.org. And I have some difficulties:

program test_MPI_RMA                                                                                                                                                       
use mpi                                                                                                                                                      
                                                                                                                                         
integer, parameter :: m=10

INTEGER i,map(m),comm,ierr,nproc,rank
REAL*8 A(m),B(m)
                                                                                                                                                          
call MPI_INIT(ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, nproc, ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)

PRINT*,"nproc=",nproc,"rank=",rank

do i=1,m
   A(i)=rank*0.1d0
   B(i)=0.d0
   if (rank==0) then
      map(i)=1
   else
      map(i)=0
   endif
   print*, 'node', rank, A(i), map(i)
enddo

call MPI_BARRIER(MPI_COMM_WORLD, ierr)

call SUM(A, B, map, m, MPI_COMM_WORLD, nproc)

call MPI_FINALIZE(ierr)

end program test_MPI_RMA



SUBROUTINE SUM(A, B, map, m, comm, p)
USE MPI

INTEGER m, map(m), comm, p, sizeofreal, win, ierr
REAL*8 A(m), B(m)

CALL MPI_TYPE_EXTENT(MPI_DOUBLE_PRECISION, sizeofreal, ierr)

CALL MPI_WIN_CREATE(B, m*sizeofreal, sizeofreal, MPI_INFO_NULL,  &
                    comm, win, ierr)

CALL MPI_WIN_FENCE(0, win, ierr)
DO i=1,m
  j = map(i)/m
  k = MOD(map(i),m)
  CALL MPI_ACCUMULATE(A(i), 1, MPI_DOUBLE_PRECISION, j, k, 1, MPI_DOUBLE_PRECISION,   &
                      MPI_SUM, win, ierr)
END DO
CALL MPI_WIN_FENCE(0, win, ierr)

CALL MPI_WIN_FREE(win, ierr)
RETURN
END

I compile this code with mpif90 and intel compiler 15.0.0 without problem. However, when I run it, I get this error:

Fatal error in MPI_Win_create: Invalid size argument in RMA call, error stack:
MPI_Win_create(201): MPI_Win_create(base=0x6a98a0, size=80, disp_unit=8, MPI_INFO_NULL, MPI_COMM_WORLD, win=0x7fff89a00384) failed
MPI_Win_create(146): Invalid size argument in RMA call (value is 80)

In others words I need help :)

Could someone tell me what is wrong in these lines of code?

Best regards,

Guillaume

[Guillaume_De_Nayer]

I reply to myself :)

I found an updated version of SUM with some difference in MPI_TYPE_GET_EXTENT. The previous error seems to be gone !?!

SUBROUTINE SUM(A, B, map, m, comm, p)
USE MPI 
INTEGER m, map(m), comm, p, win, ierr
!REAL A(m), B(m)                                                                                                                                                                                     
REAL*8 A(m), B(m)
INTEGER (KIND=MPI_ADDRESS_KIND) lowerbound, sizeofreal

CALL MPI_TYPE_GET_EXTENT(MPI_DOUBLE_PRECISION, lowerbound, sizeofreal, ierr)
CALL MPI_WIN_CREATE(B, m*sizeofreal, sizeofreal, MPI_INFO_NULL,  &
                    comm, win, ierr)

CALL MPI_WIN_FENCE(0, win, ierr)
DO i=1,m
  j = map(i)/m
  k = MOD(map(i),m)
  CALL MPI_ACCUMULATE(A(i), 1, MPI_DOUBLE_PRECISION, j, k, 1, MPI_DOUBLE_PRECISION,   &
                      MPI_SUM, win, ierr)
END DO
CALL MPI_WIN_FENCE(0, win, ierr)

CALL MPI_WIN_FREE(win, ierr)
RETURN
END

However, if someone has other simple examples of code with MPI_WIN_CREATE, please share them :)

Regards