Example using MPI (Fortran)

The following is an example for a code using MPI. This code runs successfully on HPCx. The variable me_world has been set to the rank of the task in the MPI communicator MPI_comm_world. This is done inside the subroutine make_mpi_world().

We instrumented three regions inside the code. The first region, with the label ``kern'', measures the performance of the entire kernel of the code. The second region, labeled ``communication'', measures the performance of the routine ``Halo_swap''. Finally the third region, labeled ``working'', measures the performance of the routine ``Update_image()''. The second and third region are placed inside an iteration loop. For each iteration, their counters will be increased appropriately once they are restarted. Please note that the regions ``communication'' and ``working'' are nested inside the region ``kern''.

program image_main

  use image_param
  use image_mpi_comm
  use image_work
  use image_time
  use time_tool
  
  implicit none

#include "f_hpm.h"

  real(kind=8) :: a_time, b_time, c_time 
  real(kind=8) :: s_time

  integer iter

  Call make_mpi_world()
  Call make_cart_world()
  Call edge_init('r')

  ! make sure we start together
  Call sync_this(cart_comm)

  ! set times to 0
  call zero_timers()

  call f_hpminit( me_world, "case_image_normal_nohott" )

  call f_hpmstart(1, "kern" )

  s_time = now_time()
  iterloop:  do iter = 1, Nb_iter

     call f_hpmstart( 2, "communication"  ) 
     a_time = now_time()

     Call Halo_swap(image)

     call f_hpmstop( 2 )

     call f_hpmstart( 3, "working" )
     b_time = now_time()

     Call Update_image()

     c_time = now_time()

     call f_hpmstop( 3 )
     halo_time = halo_time + b_time - a_time
     update_time = update_time + c_time - b_time

  enddo iterloop

  call f_hpmstop(1)
  elapsed_time = c_time - s_time

  Call Print_times()

  call f_hpmterminate( me_world ) 

  Call shut_mpi_world()

end