! to compile, use
! mpif90 parallelize_outer_of_triple_sum.mpi.f90
! to run, use
! mpirun -np 2 ./a.out
program exampleParallel
implicit none
include 'mpif.h'
INTEGER :: num_cpu, rank, ierr
INTEGER :: loop1Indx, loop2Indx, loop3Indx, loop4Indx
INTEGER :: loop1Limit, loop2Limit, loop3Limit, loop4Limit
INTEGER :: bound
INTEGER :: current_val, summd_val,all_summd
loop1Limit=2
loop2Limit=2
loop3Limit=1
! needs to obey loop1Limit*loop2Limit*loop3Limit=num_cpu
loop4Limit=4
current_val=0
summd_val=0
call MPI_INIT(ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, num_cpu, ierr)
write(*,*) 'Number of CPUs=',num_cpu,' My rank=',rank
loop3Indx=1
outermostloop: do while (loop3Indx.le.loop3Limit)
loop2Indx=1
do while (loop2Indx.le.loop2Limit)
bound=loop1Limit*(loop2Indx+(loop3Indx-1)*loop2Limit)
if (((bound-1).le.rank).AND.(rank.lt.bound)) then
loop1Indx=rank+1-loop2Limit*(loop2Indx-1)-loop2Limit*loop3Limit*(loop3Indx-1)
exit outermostloop
endif
loop2Indx=loop2Indx+1
enddo
loop3Indx=loop3Indx+1
enddo outermostloop
! now loop1Indx and loop2Indx are set by the rank
write(*,*) 'a=',loop1Indx,'b=',loop2Indx,'on CPU',rank
do loop4Indx = 1, loop4Limit
current_val = loop4Indx+(loop3Indx-1)*loop4Limit+(loop2Indx-1)*loop3Limit*loop4Limit+&
(loop1Indx-1)*loop2Limit*loop3Limit*loop4Limit
write(*,*) current_val, ' from CPU ', rank
summd_val = summd_val+current_val
enddo
call MPI_BARRIER(MPI_COMM_WORLD,ierr)
if (ierr.NE.MPI_SUCCESS) then
write(*,*) 'mpi_barrier failed, ierr=',ierr,'rank=',rank
stop
endif
call MPI_REDUCE(summd_val,all_summd,1,MPI_INTEGER,MPI_SUM,0,MPI_COMM_WORLD,ierr)
if (ierr.NE.MPI_SUCCESS) then
write(*,*) 'mpi_reduce validgaincount failed, ierr=',ierr,'rank=',rank
stop
endif
if (rank.eq.0) then
write(*,*) 'all summed value is ', all_summd
endif
call mpi_finalize(ierr)
end program