ecearth3/sources/tm5mp/base/domain_decomp.F90

!
#define TRACEBACK write (gol,'("in ",a," (",a,", line",i5,")")') rname, __FILE__, __LINE__; call goErr
#define IF_NOTOK_RETURN(action) if (status/=0) then; TRACEBACK; action; return; end if
#define IF_ERROR_RETURN(action) if (status> 0) then; TRACEBACK; action; return; end if
!
#define IF_NOTOK_MPI(action) if (ierr/=MPI_SUCCESS) then; TRACEBACK; action; return; end if
!
#include "tm5.inc"
!
!----------------------------------------------------------------------------
!                    TM5                                                    !
!----------------------------------------------------------------------------
!BOP
!
! !MODULE:      DOMAIN_DECOMP
!
! !DESCRIPTION: Define a distributed grid object and its methods.
!               Horizontal grid is decomposed across longitudes and latitudes.
!\\
!\\
! !INTERFACE: 
!
MODULE DOMAIN_DECOMP
  !
  ! !USES:
  !
  use grid,       only : TllGridInfo      ! Type with Lon/Lat Grid Info
  use Go,         only : goErr, goPr, gol ! go = general objects
  use dims,       only : okdebug
  use partools                            ! to include mpif.h, ierr, localComm,...
  
  IMPLICIT NONE
  PRIVATE
  !
  ! !PUBLIC MEMBER FUNCTIONS:
  !
  public :: Init_DistGrid, Done_DistGrid  ! life cycle routines
  public :: Get_DistGrid                  ! get bounds & grids
  public :: Print_DistGrid                ! print utility (debug)
  
  public :: GATHER, SCATTER, UPDATE_HALO  ! communication routines

  public :: GATHER_I_BAND,  GATHER_J_BAND ! communication routines
  public :: SCATTER_I_BAND, SCATTER_J_BAND! for distributed slices
  public :: UPDATE_HALO_JBAND             ! 
  public :: UPDATE_HALO_IBAND             ! 
  
  public :: REDUCE                        ! mimic MPI_REDUCE / MPI_ALLREDUCE
  public :: DIST_ARR_STAT                 ! basic statitics of a distributed array
  public :: testcomm                      ! test communication (debug)
  !
  ! !PUBLIC TYPES:
  !
  TYPE, PUBLIC :: DIST_GRID

     PRIVATE

     ! parameters for global domain
     integer :: im_region        ! number of longitudes
     integer :: jm_region        ! number of latitudes

     ! parameters for local domain
     integer :: i_strt          ! begin local domain longitude index
     integer :: i_stop          ! end   local domain longitude index

     integer :: i_strt_halo     ! begin halo longitude index
     integer :: i_stop_halo     ! end   halo longitude index

     integer :: j_strt          ! begin local domain latitude index
     integer :: j_stop          ! end   local domain latitude index

     integer :: j_strt_halo     ! begin halo latitude index
     integer :: j_stop_halo     ! end   halo latitude index

     type(TllGridInfo) :: lli      ! local  Lat-Lon grid info
     type(TllGridInfo) :: glbl_lli ! global Lat-Lon grid info
     type(TllGridInfo) :: lli_z    ! global meridional grid info

     integer :: neighbors(4)  ! rank of (west, north, east, south) processes

     logical :: has_south_pole  ! south pole is in local domain
     logical :: has_north_pole  ! north pole is in local domain
     logical :: has_south_boundary  ! south boundary is in local domain
     logical :: has_north_boundary  ! north boundary is in local domain
     logical :: has_west_boundary   ! west  boundary is in local domain
     logical :: has_east_boundary   ! east  boundary is in local domain
     logical :: is_periodic         ! covers [-180, 180]

     ! i_start, i_stop, j_start, j_stop of all PEs: (4,npes)
     integer, pointer :: bounds(:,:)
     
  END TYPE DIST_GRID
  !
  ! !PRIVATE DATA MEMBERS:
  !    
  character(len=*), parameter  ::  mname='Domain_Decomp_MOD_'
  !
  ! !INTERFACE:
  !
  INTERFACE Init_DistGrid
     MODULE PROCEDURE INIT_DISTGRID_FROM_REGION
     MODULE PROCEDURE INIT_DISTGRID_FROM_LLI
  END INTERFACE
     
  INTERFACE Update_halo         ! for arrays distributed accross both I and J (1st and 2nd dim)
     MODULE PROCEDURE Update_halo_2d_r
     MODULE PROCEDURE Update_halo_3d_r
     MODULE PROCEDURE Update_halo_4d_r
     MODULE PROCEDURE Update_halo_2d_i
  END INTERFACE

  INTERFACE Update_halo_jband         ! for arrays distributed accross I (1st dim)
     MODULE PROCEDURE Update_halo_jband_1d_r
     MODULE PROCEDURE Update_halo_jband_2d_r
     MODULE PROCEDURE Update_halo_jband_3d_r
     MODULE PROCEDURE Update_halo_jband_4d_r
  END INTERFACE
  
  INTERFACE Update_halo_iband         ! for arrays distributed accross J (1st dim)
     MODULE PROCEDURE Update_halo_iband_1d_r
  END INTERFACE
  
  INTERFACE Gather
     MODULE PROCEDURE gather_2d_i
     MODULE PROCEDURE gather_2d_r
     MODULE PROCEDURE gather_3d_r
     MODULE PROCEDURE gather_4d_r
     MODULE PROCEDURE gather_5d_r
  END INTERFACE
  
  INTERFACE Scatter
     MODULE PROCEDURE scatter_2d_r
     MODULE PROCEDURE scatter_3d_r
     MODULE PROCEDURE scatter_4d_r
  END INTERFACE

  INTERFACE Gather_I_Band
     MODULE PROCEDURE gather_iband_1d_r
!     MODULE PROCEDURE gather_iband_2d_r
     MODULE PROCEDURE gather_iband_3d_r
  END INTERFACE

  INTERFACE Gather_J_Band
!     MODULE PROCEDURE gather_jband_1d_r
     MODULE PROCEDURE gather_jband_2d_r
     MODULE PROCEDURE gather_jband_3d_r
     MODULE PROCEDURE gather_jband_4d_r
  END INTERFACE

  INTERFACE Scatter_I_Band
     MODULE PROCEDURE scatter_iband_1d_r
     MODULE PROCEDURE scatter_iband_2d_r
  END INTERFACE

  INTERFACE Scatter_J_Band
     MODULE PROCEDURE scatter_jband_1d_r
     MODULE PROCEDURE scatter_jband_2d_r
     MODULE PROCEDURE scatter_jband_3d_r
     MODULE PROCEDURE scatter_jband_4d_r
  END INTERFACE

  INTERFACE Reduce
     MODULE PROCEDURE reduce_2d_r
     MODULE PROCEDURE reduce_3d_r
     MODULE PROCEDURE reduce_4d_r
  END INTERFACE

  INTERFACE Dist_Arr_Stat
     MODULE PROCEDURE dist_arr_stat_2d_r
  END INTERFACE
  !
  ! !REVISION HISTORY: 
  !   01 Nov 2011 - P. Le Sager - v0
  !   13 Feb 2013 - P. Le Sager - Remove deprecated MPI_GET_EXTENT and
  !                                MPI_TYPE_HVECTOR.
  !                             - Extra garbage clean.
  !   21 Jun 2013 - P. Le Sager - Replace all MPI_SEND with MPI_SSEND in all
  !                               scatter routines to avoid buffering.
  !
  ! !REMARKS:
  ! (1) There is two categories of subroutines here : one to define the
  !     distributed grid objects, the other for communication b/w processes
  !     (update_halo for one-to-one comm, and scatter/gather/reduce for
  !     collective comm)
  !     Communications routines are for data DECOMPOSED on the domain. For
  !      more general comm, see partools module. 
  !     
  ! (2) GATHER & SCATTER :
  !     - global arrays have to really be global on root only, and can be 
  !        (1,1,..) on other processes.
  !     - global arrays are without halo.
  !     - if not using MPI, gather and scatter just copy arrays, so it may be
  !        better to not call them to save memory in that case.
  !
  ! (3) Be careful when passing a slice (or a pointer to a slice) as argument:
  !     
  !       Passing a subarray will *NOT* work most of the time, unless it is a
  !       slice on the last dimension(s). This will give erroneous results:
  !       
  !           allocate( gbl_arr(-3:imr, 1:jmr ))
  !           call gather( dgrid, local_arr, gbl_arr(1:imr,:), halo, status)
  !       
  !       but passing full size array will work:
  !       
  !           allocate( gbl_arr(-3:imr, 1:jmr ))
  !           allocate( temp(1:imr,1:jmr) )
  !           call gather( dgrid, local_arr, temp, halo, status)
  !           gbl_arr(1:imr,:) = temp
  !
  ! (4) Scatter[Gather]_I[J]_band are communications between a row or column
  !      of processors and the root processor (optionally the row_root in few
  !       cases).  
  !EOP
  !------------------------------------------------------------------------

CONTAINS
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    DISTGRID_RANGE
  !
  ! !DESCRIPTION: Give range of indices covered by rank when using nprocs.
  !               This is used for one dimension, and thus called twice.
  !               Distribution is done evenly. Eg: it will distribute 11 boxes
  !               across 3 processes as 4,4,3 on each pe.
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE DISTGRID_RANGE(ij, rank, nprocs, istart, iend)
    !
    ! !INPUT PARAMETERS:
    !
    integer, intent(in) :: ij      ! defines range (1,..,ij) to be distributed
    integer, intent(in) :: rank    ! current process, in (0,.., nprocs-1)
    integer, intent(in) :: nprocs  ! total # of processes
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out):: istart, iend    ! index range covered by rank
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    integer :: iwork1, iwork2

    iwork1 = ij/nprocs
    iwork2 = mod(ij,nprocs)

    istart = rank * iwork1 + 1 + min(rank, iwork2)
    iend = istart + iwork1 - 1
    if (iwork2 > rank) iend = iend + 1

  END SUBROUTINE DISTGRID_RANGE
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    INIT_DISTGRID_FROM_REGION
  !
  ! !DESCRIPTION: initialize a distributed grid object for a TM5 region
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE INIT_DISTGRID_FROM_REGION( DistGrid, region, rank, nplon, nplat, halo, status)
    !
    ! !USES:
    !    
    use grid, only : init
    use dims, only : xbeg, xend, dx, im, xcyc
    use dims, only : ybeg, yend, dy, jm, touch_np, touch_sp
    !
    ! !RETURN VALUE:
    !
    type(dist_grid), intent(inout) :: DistGrid
    !
    ! !INPUT PARAMETERS:
    !
    integer, intent(in)           :: region   ! TM5 region number
    integer, intent(in)           :: rank     ! current process in (0,.., nprocs-1)
    integer, intent(in)           :: nplon    ! number of processes along longitude
    integer, intent(in)           :: nplat    ! number of processes along latitude
    integer, intent(in), optional :: halo     ! halo size (default=0)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS:
    ! 
    ! (1) Ideally: to call with WIDTH=<halo used in the code>... but not the
    ! same halo is used for all data sets. Then, in many subroutines, halo has
    ! to be independent of the halo carried by the distributed grid. We could
    ! simplify things by using a halo of 2 for all distributed data, or
    ! disregard halo in the dist_grid, but we just keep it general (as is).
    !
    ! (2) npe_lat/lon are also available thru partools, but may not have been
    ! initialized yet, so we keep the nplon/nplat inputs here. Could do
    ! without though...
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'init_distgrid'

    integer :: start, end, iwidth, i, iml, jml, iwork(2), lshape(4)
    real    :: yb, xb, dlon, dlat
    character(len=39) :: name

    ! for pretty print
    integer, parameter :: maxrow=5
    integer, parameter :: maxcol=5
    integer            :: i1, j1, j2, k, work
    character(len=3)   :: str1

    ! for virtual topology
    integer, allocatable :: dist_map(:,:)  ! 2D process topology (np_lon, np_lat)
    integer, allocatable :: dist_mapw(:,:) ! 2D process topology with neighbors (np_lon+1, np_lat+1)
    integer :: rank_lon ! index of current process in azimuthal set (1,..,np_lon)
    integer :: rank_lat ! index of current process in meridional set (1,..,np_lat)

    !---------------------------------------------
    ! 2D process topology, and mapping 1D <-> 2D
    !---------------------------------------------
    ! besides the periodicity, topology is independent of the region
    allocate(dist_map(nplon, nplat))
    npes = nplat*nplon
    dist_map = reshape( (/ (i,i=0,npes-1) /), shape=(/ nplon, nplat /))
        
    ! store
    iwork    = maxloc(dist_map, mask=dist_map.eq.rank)
    rank_lon = iwork(1)   ! in [1..nplon]
    rank_lat = iwork(2)   ! in [1..nplat]

    ! Neighbors = 2d process map with extra neighbors
    allocate(dist_mapw(0:nplon+1, 0:nplat+1))
    dist_mapw = MPI_PROC_NULL
    dist_mapw(1:nplon,1:nplat) = dist_map

    ! East-West periodicity
    DistGrid%is_periodic =.false.
    if (xcyc(region)==1) then
       dist_mapw(      0,1:nplat) = dist_map(nplon,:)
       dist_mapw(nplon+1,1:nplat) = dist_map(1,:)
       DistGrid%is_periodic =.true.
    end if

    DistGrid%neighbors = (/ dist_mapw(rank_lon-1, rank_lat  ), & ! west
                            dist_mapw(rank_lon,   rank_lat+1), & ! north
                            dist_mapw(rank_lon+1, rank_lat  ), & ! east
                            dist_mapw(rank_lon,   rank_lat-1) /) ! south

    !---------------------------------------------
    ! fill in distributed grid info
    !---------------------------------------------
    ! region boundaries
    DistGrid%has_south_boundary = (rank_lat == 1    )
    DistGrid%has_north_boundary = (rank_lat == nplat)
    DistGrid%has_west_boundary  = (rank_lon == 1    )
    DistGrid%has_east_boundary  = (rank_lon == nplon)

    ! poles
    DistGrid%has_south_pole = DistGrid%has_south_boundary .and. (touch_sp(region) == 1)
    DistGrid%has_north_pole = DistGrid%has_north_boundary .and. (touch_np(region) == 1)
        
    ! max halo that will be used in the code
    iwidth=0
    if (present(halo)) iwidth=halo

    ! index ranges covered by current process
    call DistGrid_range(im(region), rank_lon-1, nplon, start, end)

    ! check we are within the limit, i.e. we must have #boxes >= halo and at least 1.
    if ( (end-start+1) < max(1,iwidth) ) then
       write(gol,*)" Too much processors in X (longitude) direction! ", nplon, iwidth, start, end
       call goErr
       status=1; TRACEBACK; return
    end if
    
    DistGrid%im_region   = im(region)
    DistGrid%i_strt      = start
    DistGrid%i_stop      = end
    DistGrid%i_strt_halo = start-iwidth
    DistGrid%i_stop_halo = end+iwidth

! To think about it when/if needed:    
!    if (DistGrid%has_north_pole) DistGrid%i_stop_halo = end
!    if (DistGrid%has_south_pole) DistGrid%i_strt_halo = start

    call DistGrid_range(jm(region), rank_lat-1, nplat, start, end)

    if ( (end-start+1) < max(1,iwidth) ) then
       write(gol,*)" Too much processors in Y (latitude) direction! ", nplat, iwidth, start, end
       call goErr
       status=1; TRACEBACK; return
    end if

    DistGrid%jm_region   = jm(region)
    DistGrid%j_strt      = start
    DistGrid%j_stop      = end
    DistGrid%j_strt_halo = start-iwidth
    DistGrid%j_stop_halo = end+iwidth
    
#ifdef parallel_cplng
    if ( modulo(im(region),nplon) /= 0) then
       write(gol,'("number of X processors ",i3," does not divide evenly the number of grid boxes",i3)') &
            nplon, im(region) ; call goErr
       status=1; TRACEBACK; return
    endif
    if ( modulo(jm(region),nplat) /= 0) then
       write(gol,'("number of Y processors ",i3," does not divide evenly the number of grid boxes ",i3)') &
            nplat, jm(region) ; call goErr
       status=1; TRACEBACK; return
    endif
#endif    
    
    !---------------------------------------------
    ! geophysical grids
    !---------------------------------------------
    ! grid spacing: 
    dlon = real(xend(region)-xbeg(region))/im(region)
    dlat = real(yend(region)-ybeg(region))/jm(region)

    !------ local
    iml = DistGrid%i_stop - DistGrid%i_strt + 1
    jml = DistGrid%j_stop - DistGrid%j_strt + 1

    xb = xbeg(region) + ( DistGrid%i_strt - 0.5 ) * dlon
    yb = ybeg(region) + ( DistGrid%j_strt - 0.5 ) * dlat

    write(name, '("distributed grid on process ", i5.5)') rank
    
    call INIT( DistGrid%lli, xb, dlon, iml, yb, dlat, jml, status, name=name )
    IF_NOTOK_RETURN(status=1)

    !------ global
    xb = xbeg(region) + 0.5 * dlon
    yb = ybeg(region) + 0.5 * dlat

    write(name, '("global grid on process ", i5.5)') rank

    call INIT( DistGrid%glbl_lli, xb, dlon, im(region), yb, dlat, jm(region), status, name=name )
    IF_NOTOK_RETURN(status=1)
    
    !------ global meridional grid
    name = "merid " // trim(name)

    call INIT( DistGrid%lli_z, 0.0, 360., 1, yb, dlat, jm(region), status, name=name )
    IF_NOTOK_RETURN(status=1)

    !---------------------------------------------
    ! store shapes info of all PE on all PE (used only on root so far, but may
    ! become useful if we gather/scatter to/from non-root in the future)
    !---------------------------------------------
#ifdef MPI    
    allocate(DistGrid%bounds(4,0:npes-1))

    lshape = (/ DistGrid%i_strt, DistGrid%i_stop, DistGrid%j_strt, DistGrid%j_stop /)

    call MPI_ALLGATHER(lshape, 4, MPI_INTEGER, DistGrid%bounds, 4, MPI_INTEGER, localComm, ierr)
#endif

    !---------------------------------------------
    ! PRINT (Debug) 2D process topology
    !---------------------------------------------
    if(okdebug)then

       write(gol,*) '------------- world view ----------------------' ; call goPr

       write(gol, '("process #", i5, " out of ", i5)') myid, npes ; call goPr
       write(gol, '("2D rank = [",i4,", ",i4,"]")') rank_lon, rank_lat ; call goPr

       i1=min(maxcol,nplon)

       str1='...'
       if (i1==nplon) str1=''

       work=nplat/2
       j1=min(work, maxrow)
       j2=max(nplat-maxrow+1, work+1)

       do k=nplat,j2,-1
          write(gol,*) dist_map(1:i1,k),str1 ; call goPr
       enddo
       if ((j2-1) /= j1) write(gol,*)"     ..." ; call goPr
       do k=j1,1,-1
          write(gol,*) dist_map(1:i1,k),str1 ; call goPr
       enddo

       write(gol,*) ''
       write(gol,*) '-----------------------------------------------' ; call goPr

    end if
    
    ! Done
    deallocate(dist_map)
    deallocate(dist_mapw)
    status = 0

  END SUBROUTINE INIT_DISTGRID_FROM_REGION
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    INIT_DISTGRID_FROM_LLI
  !
  ! !DESCRIPTION: initialize a distributed grid object from a lli object
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE INIT_DISTGRID_FROM_LLI( DistGrid, lli, rank, halo, status)
    !
    ! !USES:
    !    
    use grid, only : init, assignment(=)
    !
    ! !INPUT/OUTPUT PARAMETERS:
    !
    type(dist_grid), intent(inout) :: DistGrid
    !
    ! !INPUT PARAMETERS:
    !
    type(TllGridInfo),  intent(in) :: lli
    integer,            intent(in) :: rank  ! current process in (0,.., nprocs-1)
    integer,  optional, intent(in) :: halo  ! halo size (default=0)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   18 Nov 2013 - Ph. Le Sager - v0, adapted from init_distgrid_from_region
    !
    ! !REMARKS:
    ! 
    ! (1) See INIT_DISTGRID_FROM_REGION for comment about halo input.
    ! (2) Uses npe_lat/lon from partools module : TM5_MPI_Init2 must have been
    !      called before. 
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'init_distgrid_from_lli'

    integer :: start, end, iwidth, i, iml, jml, iwork(2), lshape(4)
    real    :: yb, xb, dlon, dlat
    character(len=39) :: name

    integer  :: nplon    ! number of processes along longitude
    integer  :: nplat    ! number of processes along latitude

    ! for pretty print
    integer, parameter :: maxrow=5
    integer, parameter :: maxcol=5
    integer            :: i1, j1, j2, k, work
    character(len=3)   :: str1

    ! for virtual topology
    integer, allocatable :: dist_map(:,:)  ! 2D process topology (np_lon, np_lat)
    integer, allocatable :: dist_mapw(:,:) ! 2D process topology with neighbors (np_lon+1, np_lat+1)
    integer :: rank_lon ! index of current process in azimuthal set (1,..,np_lon)
    integer :: rank_lat ! index of current process in meridional set (1,..,np_lat)

    !---------------------------------------------
    ! test for valid input
    !---------------------------------------------
    if (.not.(associated( lli%lon_deg ))) then
       write(gol,*)" The LLI object has not been initialized.   " ; call goErr
       write(gol,*)" A distributed grid for MPI cannot be build." ; call goErr
       status=1; TRACEBACK; return
    end if

    nplat = npe_lat
    nplon = npe_lon
    
    !---------------------------------------------
    ! 2D process topology, and mapping 1D <-> 2D
    !---------------------------------------------
    ! besides the periodicity, topology is independent of the region
    allocate(dist_map(nplon, nplat))
    npes = nplat*nplon
    dist_map = reshape( (/ (i,i=0,npes-1) /), shape=(/ nplon, nplat /))

    ! store
    iwork    = maxloc(dist_map, mask=dist_map.eq.rank)
    rank_lon = iwork(1)   ! in [1..nplon]
    rank_lat = iwork(2)   ! in [1..nplat]

    ! Neighbors = 2d process map with extra neighbors
    allocate(dist_mapw(0:nplon+1, 0:nplat+1))
    dist_mapw = MPI_PROC_NULL
    dist_mapw(1:nplon,1:nplat) = dist_map

    ! East-West periodicity
    DistGrid%is_periodic =.false.

    if ( lli%dlon_deg*lli%im == 360. ) then 
       dist_mapw(      0,1:nplat) = dist_map(nplon,:)
       dist_mapw(nplon+1,1:nplat) = dist_map(1,:)
       DistGrid%is_periodic =.true.
    end if

    DistGrid%neighbors = (/ dist_mapw(rank_lon-1, rank_lat  ), & ! west
                            dist_mapw(rank_lon,   rank_lat+1), & ! north
                            dist_mapw(rank_lon+1, rank_lat  ), & ! east
                            dist_mapw(rank_lon,   rank_lat-1) /) ! south

    !---------------------------------------------
    ! fill in distributed grid info
    !---------------------------------------------
    ! region boundaries
    DistGrid%has_south_boundary = (rank_lat == 1    )
    DistGrid%has_north_boundary = (rank_lat == nplat)
    DistGrid%has_west_boundary  = (rank_lon == 1    )
    DistGrid%has_east_boundary  = (rank_lon == nplon)

    ! poles
    DistGrid%has_south_pole = DistGrid%has_south_boundary .and. (lli%blat_deg(0) == -90.0)
    DistGrid%has_north_pole = DistGrid%has_north_boundary .and. (lli%blat_deg(lli%jm) == 90.0)

    ! max halo that will be used in the code
    iwidth=0
    if (present(halo)) iwidth=halo

    ! index ranges covered by current process
    call DistGrid_range(lli%im, rank_lon-1, nplon, start, end)

    ! check we are within the limit, i.e. we must have #boxes >= halo and at least 1.
    if ( (end-start+1) < max(1,iwidth) ) then
       write(gol,*)" Too much processors in X (longitude) direction:", nplon, iwidth, start, end
       call goErr
       status=1; TRACEBACK; return
    end if

    DistGrid%im_region   = lli%im
    DistGrid%i_strt      = start
    DistGrid%i_stop      = end
    DistGrid%i_strt_halo = start-iwidth
    DistGrid%i_stop_halo = end+iwidth

    ! To think about it when/if needed:    
    !    if (DistGrid%has_north_pole) DistGrid%i_stop_halo = end
    !    if (DistGrid%has_south_pole) DistGrid%i_strt_halo = start

    call DistGrid_range(lli%jm, rank_lat-1, nplat, start, end)

    if ( (end-start+1) < max(1,iwidth) ) then
       write(gol,*)" Too much processors in Y (latitude) direction:", nplat, iwidth, start, end
       call goErr
       status=1; TRACEBACK; return
    end if

    DistGrid%jm_region   = lli%jm
    DistGrid%j_strt      = start
    DistGrid%j_stop      = end
    DistGrid%j_strt_halo = start-iwidth
    DistGrid%j_stop_halo = end+iwidth

    !---------------------------------------------
    ! geophysical grids
    !---------------------------------------------

    !------ tile
    iml = DistGrid%i_stop - DistGrid%i_strt + 1
    jml = DistGrid%j_stop - DistGrid%j_strt + 1

    xb = lli%lon_deg(1) + ( DistGrid%i_strt - 1 ) * lli%dlon_deg
    yb = lli%lat_deg(1) + ( DistGrid%j_strt - 1 ) * lli%dlat_deg

    write(name, '("distributed grid on process ", i5.5)') rank

    call INIT( DistGrid%lli, xb, lli%dlon_deg, iml, yb, lli%dlat_deg, jml, status, name=name )
    IF_NOTOK_RETURN(status=1)

    !------ world

    DistGrid%glbl_lli = lli
    
    !------ world meridional grid
    name = "merid "

    call INIT( DistGrid%lli_z, 0.0, 360., 1, yb, lli%dlat_deg, lli%jm, status, name=name )
    IF_NOTOK_RETURN(status=1)

    !---------------------------------------------
    ! store shapes info of all PE on all PE (used only on root so far, but may
    ! become useful if we gather/scatter to/from non-root in the future)
    !---------------------------------------------
#ifdef MPI    
    allocate(DistGrid%bounds(4,0:npes-1))

    lshape = (/ DistGrid%i_strt, DistGrid%i_stop, DistGrid%j_strt, DistGrid%j_stop /)

    call MPI_ALLGATHER(lshape, 4, MPI_INTEGER, DistGrid%bounds, 4, MPI_INTEGER, localComm, ierr)
#endif

    !---------------------------------------------
    ! PRINT (Debug) 2D process topology
    !---------------------------------------------
    if(okdebug)then

       write(gol,*) '------------- world view ----------------------' ; call goPr

       write(gol, '("process #", i5, " out of ", i5)') myid, npes ; call goPr
       write(gol, '("2D rank = [",i4,", ",i4,"]")') rank_lon, rank_lat ; call goPr

       i1=min(maxcol,nplon)

       str1='...'
       if (i1==nplon) str1=''

       work=nplat/2
       j1=min(work, maxrow)
       j2=max(nplat-maxrow+1, work+1)

       do k=nplat,j2,-1
          write(gol,*) dist_map(1:i1,k),str1 ; call goPr
       enddo
       if ((j2-1) /= j1) write(gol,*)"     ..." ; call goPr
       do k=j1,1,-1
          write(gol,*) dist_map(1:i1,k),str1 ; call goPr
       enddo

       write(gol,*) ''
       write(gol,*) '-----------------------------------------------' ; call goPr

    end if

    ! Done
    deallocate(dist_map)
    deallocate(dist_mapw)
    status = 0

  END SUBROUTINE INIT_DISTGRID_FROM_LLI
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    DONE_DISTGRID
  !
  ! !DESCRIPTION: deallocate distributed grid object elements
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE DONE_DISTGRID( DistGrid, STATUS )
    !
    ! !USES:
    !
    use grid, only : done
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(inout) :: DistGrid
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'Done_Distgrid'

    call done(DistGrid%lli, status)
    IF_NOTOK_RETURN(status=1)

    call done(DistGrid%lli_z, status)
    IF_NOTOK_RETURN(status=1)

    call done(DistGrid%glbl_lli, status)
    IF_NOTOK_RETURN(status=1)
    
    if (associated(DistGrid%bounds)) deallocate(DistGrid%bounds)
    
    status=0 

  END SUBROUTINE DONE_DISTGRID
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GET_DISTGRID
  !
  ! !DESCRIPTION: provide quick access to object elements (bounds and grids),
  !               while keeping them private.
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GET_DISTGRID(DistGrid, &
                          i_strt, i_stop, i_strt_halo, i_stop_halo, &
                          j_strt, j_stop, j_strt_halo, j_stop_halo, &
                          i_wrld, j_wrld,                           &
                          hasSouthPole,   hasNorthPole,             & 
                          hasSouthBorder, hasNorthBorder,           & 
                          hasEastBorder,  hasWestBorder,            & 
                          lli,    lli_z,  global_lli                )
    !
    ! !USES:
    !
    use grid, only : assignment(=)      ! to copy lli
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),  intent(in) :: DistGrid
    integer, optional :: i_strt, i_stop, i_strt_halo, i_stop_halo
    integer, optional :: j_strt, j_stop, j_strt_halo, j_stop_halo
    integer, optional :: i_wrld, j_wrld
    logical, optional :: hasSouthPole,   hasNorthPole
    logical, optional :: hasSouthBorder, hasNorthBorder
    logical, optional :: hasWestBorder,  hasEastBorder
    type(TllGridInfo), optional :: lli          ! local Lat-Lon grid info
    type(TllGridInfo), optional :: global_lli   ! global Lat-Lon grid info
    type(TllGridInfo), optional :: lli_z        ! global zonal grid info
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    if (present(i_strt))      i_strt = DistGrid%i_strt
    if (present(i_stop))      i_stop = DistGrid%i_stop
    if (present(i_strt_halo)) i_strt_halo = DistGrid%i_strt_halo
    if (present(i_stop_halo)) i_stop_halo = DistGrid%i_stop_halo

    if (present(j_strt))      j_strt = DistGrid%j_strt
    if (present(j_stop))      j_stop = DistGrid%j_stop
    if (present(j_strt_halo)) j_strt_halo = DistGrid%j_strt_halo
    if (present(j_stop_halo)) j_stop_halo = DistGrid%j_stop_halo

    if (present(i_wrld))      i_wrld = DistGrid%im_region
    if (present(j_wrld))      j_wrld = DistGrid%jm_region

    if (present(hasSouthPole)) hasSouthPole = DistGrid%has_south_pole
    if (present(hasNorthPole)) hasNorthPole = DistGrid%has_north_pole
    
    if (present(hasSouthBorder)) hasSouthBorder = DistGrid%has_south_boundary
    if (present(hasNorthBorder)) hasNorthBorder = DistGrid%has_north_boundary
    if (present(hasEastBorder )) hasEastBorder  = DistGrid%has_east_boundary
    if (present(hasWestBorder )) hasWestBorder  = DistGrid%has_west_boundary

    if (present(lli))         lli        = DistGrid%lli
    if (present(global_lli))  global_lli = DistGrid%glbl_lli
    if (present(lli_z))       lli_z      = DistGrid%lli_z
    
  END SUBROUTINE GET_DISTGRID
  !EOC
  
#ifdef MPI  /* MPI TYPES */

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GET_HALO_TYPE
  !
  ! !DESCRIPTION: Returns derived MPI types needed for halo communications,
  !               and the start indices for the send & receive communications.
  !               Four of each are returned, one for each side of the domain,
  !               in the following order: West, North, East, South.
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GET_HALO_TYPE( DistGrid, arr_shape, halo, datatype, srtype, ijsr, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    integer,         intent(in) :: arr_shape(:) ! shape of local array
    integer,         intent(in) :: halo         ! halo size
    integer,         intent(in) :: datatype     ! basic MPI datatype    
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out) :: srtype(4)      ! derived MPI datatypes for send/recv
    integer, intent(out) :: ijsr(4,4)      ! start indices for send/recv
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS:
    !  (1) Not tested on all imaginable cases, but should work with any size
    !       GE 2, and any of the basic numerical MPI_xxx datatype
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'get_halo_type'

    integer :: nsslice, weslice, nstype, wetype    ! MPI derived datatypes
    integer :: n, m, hstride                       ! sizes to build datatypes
    integer :: ndims, sz, i0, j0, i1, j1
    integer(kind=MPI_ADDRESS_KIND) :: sizeoftype, lb, vstride

    sz = size(arr_shape)

    ! collapse third and above dimensions
    ndims = 1
    if (sz > 2) ndims = product(arr_shape(3:))

    ! strides
    CALL MPI_TYPE_GET_EXTENT( datatype, lb, sizeoftype, ierr) 
    IF_NOTOK_MPI(status=1)

    hstride = arr_shape(1)
    vstride = arr_shape(1)*arr_shape(2)*sizeoftype
    
    ! size of data slice to carry 
    n = DistGrid%I_STOP - DistGrid%I_STRT + 1
    m = DistGrid%J_STOP - DistGrid%J_STRT + 1

    ! Type for North and South halo regions
    ! --------------------------------------
    ! halo lines of lenght N, separated by hstride
    call MPI_TYPE_VECTOR (halo, n, hstride, datatype, nsslice, ierr)
    IF_NOTOK_MPI(status=1)

    ! stack 3rd (and above) dimension if any
    if (ndims == 1) then
       nstype = nsslice
    else
       ! note : also works with ndims=1
       call MPI_TYPE_CREATE_HVECTOR (ndims, 1, vstride, nsslice, nstype, ierr)
       IF_NOTOK_MPI(status=1)
       
       call MPI_TYPE_FREE(nsslice, ierr)
       IF_NOTOK_MPI(status=1)
    end if

    call MPI_TYPE_COMMIT (nstype, ierr)
    IF_NOTOK_MPI(status=1)

    ! Type for West and East halo regions
    ! ------------------------------------
    ! M lines of lenght 'halo', separated by hstride
    call MPI_TYPE_VECTOR (m, halo, hstride, datatype, weslice, ierr)
    IF_NOTOK_MPI(status=1)

    ! stack 3rd (and above) dimension
    if (ndims == 1) then
       wetype = weslice
    else
       ! note : also works with ndims=1
       call MPI_TYPE_CREATE_HVECTOR (ndims, 1, vstride, weslice, wetype, ierr)   
       IF_NOTOK_MPI(status=1)

       call MPI_TYPE_FREE(weslice, ierr)
       IF_NOTOK_MPI(status=1)
    end if
    
    call MPI_TYPE_COMMIT (wetype, ierr)
    IF_NOTOK_MPI(status=1)

    ! Buffers anchors
    ! ----------------
    ! Assuming that neighbors are stored in (West, North, East, South) order
    i0  = DistGrid%i_strt
    j0  = DistGrid%j_strt        
    i1  = DistGrid%i_stop
    j1  = DistGrid%j_stop    

    srtype    = (/  wetype,    nstype,    wetype,  nstype /)
    ijsr(:,1) = (/      i0,        i0, i1+1-halo,      i0 /) ! i-start location of buffer to send
    ijsr(:,2) = (/      j0, j1-halo+1,        j0,      j0 /) ! j-start location of buffer to send
    ijsr(:,3) = (/ i0-halo,        i0,      i1+1,      i0 /) ! i-start location of buffer to receive
    ijsr(:,4) = (/      j0,      j1+1,        j0, j0-halo /) ! j-start location of buffer to receive

    status=0
    
  END SUBROUTINE GET_HALO_TYPE
  !EOC
  
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GET_INTERIOR_TYPE
  !
  ! !DESCRIPTION: Returns derived MPI types that describe the interior domains
  !               needed for collective communications.
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GET_INTERIOR_TYPE( DistGrid, shp, datatype, linterior, ginterior, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    integer,         intent(in) :: shp(:)     ! shape of local array
    integer,         intent(in) :: datatype   ! basic MPI datatype
    !
    ! !OUTPUT PARAMETERS:
    !
    ! derived MPI datatypes describing distributed interior domains:
    integer, intent(out) :: ginterior(npes-1) ! within global array (used by root, as many as NPES-1)
    integer, intent(out) :: linterior         ! within local array (used by non-root)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS:
    !  (1) input must be checked before by calling CHECK_DIST_ARR first
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'get_interior_type'

    integer :: gslice, lslice     ! intermediate datatypes
    integer :: n, m               ! sizes to build datatypes
    integer :: hlstride, hgstride ! strides to build datatypes
    integer :: stack, sz, klm
    integer(kind=MPI_ADDRESS_KIND) :: sizeoftype, lb, vlstride, vgstride

    ! init : number of dimensions, default value
    sz = size(shp)
    ginterior = MPI_DATATYPE_NULL
    linterior = MPI_DATATYPE_NULL

    ! collapse third and above dimensions
    stack = 1
    if (sz > 2) stack = product(shp(3:))
    
    ! size of data slice to carry 
    n  = DistGrid%I_STOP - DistGrid%I_STRT + 1
    m  = DistGrid%J_STOP - DistGrid%J_STRT + 1

    CALL MPI_TYPE_GET_EXTENT( datatype, lb, sizeoftype, ierr) 
    IF_NOTOK_MPI(status=1)
    
    ! horizontal global stride (in data)
    hgstride = DistGrid%im_region         

    ! vertical global stride (in byte)
    vgstride = DistGrid%im_region * DistGrid%jm_region * sizeoftype

    ! local strides (may be different than n and n*m because of halo)
    hlstride = shp(1)                    ! in data
    vlstride = shp(1)*shp(2)*sizeoftype  ! in byte

    if ( isRoot ) then

       do klm=1,npes-1

          ! horizontal chunk is M lines of lenght N, separated by a global
          ! stride 
          n = DistGrid%bounds(2,klm) - DistGrid%bounds(1,klm) + 1
          m = DistGrid%bounds(4,klm) - DistGrid%bounds(3,klm) + 1

          call MPI_TYPE_VECTOR (m, n, hgstride, datatype, gslice, ierr)
          IF_NOTOK_MPI(status=1)
          
          ! stack 3rd and above dimensions
          if (stack == 1) then
             ginterior(klm) = gslice
          else             
             ! note : also works with stack=1
             call MPI_TYPE_CREATE_HVECTOR(stack, 1, vgstride, gslice, ginterior(klm), ierr)
             IF_NOTOK_MPI(status=1)

             call MPI_TYPE_FREE(gslice, ierr)
             IF_NOTOK_MPI(status=1)
          end if
          
          call MPI_TYPE_COMMIT (ginterior(klm), ierr)
          IF_NOTOK_MPI(status=1)
          
       end do

    else
    
       ! local interior is basically M lines of lenght N, separated by a local
       ! stride 
       call MPI_TYPE_VECTOR (m, n, hlstride, datatype, lslice, ierr)
       IF_NOTOK_MPI(status=1)
       
       ! stack 3rd and above dimensions
       if (stack == 1) then
          linterior = lslice
       else
          ! note : also works with stack=1
          call MPI_TYPE_CREATE_HVECTOR (stack, 1, vlstride, lslice, linterior, ierr)
          IF_NOTOK_MPI(status=1)

          call MPI_TYPE_FREE(lslice, ierr)
          IF_NOTOK_MPI(status=1)
       end if
       
       call MPI_TYPE_COMMIT (linterior, ierr)
       IF_NOTOK_MPI(status=1)
       
    end if

    status=0
    
  END SUBROUTINE GET_INTERIOR_TYPE
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    FREE_DERIVED_TYPE
  !
  ! !DESCRIPTION: free all MPI derived datatypes in a vector
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE FREE_DERIVED_TYPE( datatype )
    !
    ! !INPUT/OUTPUT PARAMETERS:
    !
    integer, intent(inout) :: datatype(:)      ! set of derived MPI datatypes
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC

    integer :: i, j
    integer :: res(size(datatype))  ! hold unique elements 
    integer :: k                    ! number of unique elements
    
    ! Get list of unique handle(s)
    ! ----------------------------
    k = 1
    res(1) = 1
    
    outer: do i=2,size(datatype)

       ! look for a match
       do j=1,k
          if (datatype(res(j)) == datatype(i)) cycle outer  ! match
       end do
       
       ! no match : add it to the list
       k = k + 1
       res(k) = i
       
    end do outer

    ! Free handle(s)
    ! ---------------------------
    do i=1,k
       if (datatype(res(i)) /= MPI_DATATYPE_NULL) &
            call MPI_TYPE_FREE(datatype(res(i)), ierr)
    end do

  END SUBROUTINE FREE_DERIVED_TYPE
  !EOC

#endif /* MPI TYPES */
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    CHECK_DIST_ARR
  !
  ! !DESCRIPTION: Check that the shape of a local array correspond to an array
  !               distributed on current process. This check is done on the
  !               first 2 dimensions only, along which the domain
  !               decomposition is done. 
  !               
  !               Optionally: check shape of a global array. If arrays are 3D
  !               or more, the 3rd and above dimensions of local and global
  !               arrays are also compared. (This becomes "2D or more" for I-
  !               and J-Slices.)
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE CHECK_DIST_ARR( DistGrid, shp, halo, status, glbl_shp, jband, iband, has_global )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    integer,           intent(in) :: shp(:)       ! shape of local array
    integer,           intent(in) :: halo         ! halo size
    !
    integer, intent(in), optional :: glbl_shp(:)  ! shape of global array
    logical, intent(in), optional :: jband, iband ! is it a zonal or meridional slice?
    logical, intent(in), optional :: has_global   ! current proc hold global array (default is root)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !   
    ! !REMARKS: i-band refers to a unique i value, i.e. a meridional dataset.
    !           j-band refers to a unique j value, i.e. a zonal dataset.
    ! 
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'check_dist_arr '
    integer :: n, m, sz
    integer, allocatable :: glbsz(:)
    logical :: x_jband, x_iband, hold_global

    status = 0

    ! check inputs
    x_jband=.false.
    x_iband=.false.
    if(present(iband))x_iband=iband
    if(present(jband))x_jband=jband
    if(x_iband.and.x_jband)then
       write (gol,*) "CHECK_DIST_ARR: cannot have **both** I- and J-Slices"  ; call goErr
       TRACEBACK; status=1; return
    end if

    ! by default global array is expected on root
    hold_global = isRoot
    if (present(has_global)) hold_global=has_global
        
    ! required size w/o halo  
    n = DistGrid%I_STOP - DistGrid%I_STRT + 1
    m = DistGrid%J_STOP - DistGrid%J_STRT + 1

    ! check shape of array
    if (x_iband) then
       if (shp(1) /= m+2*halo) then
          write (gol,*) "CHECK_DIST_ARR: local array shape does not conform" ; call goErr
          write (gol,'("  local array : ", i4)') shp(1)                      ; call goErr
          write (gol,'("  should be   : ", i4)') m+2*halo                    ; call goErr
          write (gol,'("  with J-stop : ", i4)') DistGrid%J_STOP             ; call goErr
          write (gol,'("       J-start: ", i4)') DistGrid%J_STRT             ; call goErr
          TRACEBACK; status=1; return
       end if
    else if (x_jband) then
       if (shp(1) /= n+2*halo) then
          write (gol,*) "CHECK_DIST_ARR: local array shape does not conform" ; call goErr
          write (gol,'("  local array : ",2i4)') shp(1)                      ; call goErr    
          write (gol,'("  should be   : ",2i4)') n+2*halo                    ; call goErr           
          write (gol,'("  with J-stop : ", i4)') DistGrid%I_STOP             ; call goErr
          write (gol,'("       J-start: ", i4)') DistGrid%I_STRT             ; call goErr
          TRACEBACK; status=1; return
       end if
    else
       if ((shp(1) /= n+2*halo).or.(shp(2) /= m+2*halo)) then
          write (gol,*) "CHECK_DIST_ARR: local array shape does not conform" ; call goErr
          write (gol,'("  local array : ",2i4)') shp(1:2)                    ; call goErr    
          write (gol,'("  should be   : ",2i4)') n+2*halo, m+2*halo          ; call goErr
          write (gol,'("  w/ I/J-stop : ", i4)') DistGrid%I_STOP, DistGrid%J_STOP ; call goErr
          write (gol,'("     I/J-start: ", i4)') DistGrid%I_STRT, DistGrid%J_STRT ; call goErr
          TRACEBACK; status=1; return
       end if
    end if

    ! check shape of global array on root
    if (present(glbl_shp) .and. hold_global) then

       sz = size(shp)
       
       if (sz /= size(glbl_shp)) then
          write (gol,'("CHECK_DIST_ARR : global and local arrays have different rank:")') ; call goErr
          write (gol,'("  local rank  : ", i4)') sz                                       ; call goErr
          write (gol,'("  global rank : ", i4)') size(glbl_shp)                           ; call goErr
          TRACEBACK; status=1; return          
       end if

       allocate(glbsz(sz))
       glbsz = shp
       if (x_iband) then
          glbsz(1) = DistGrid%jm_region
       else if (x_jband) then
          glbsz(1) = DistGrid%im_region
       else
          glbsz(1:2) = (/ DistGrid%im_region, DistGrid%jm_region /)
       end if
       
       if ( any (glbl_shp /= glbsz) ) then
          write (gol,'("CHECK_DIST_ARR : global array has wrong shape:")')   ; call goErr
          write (gol,*) "                       shape(global) : ", glbl_shp  ; call goErr
          write (gol,*) "  im [and/or] jm_region/-extra dims- : ", glbsz     ; call goErr
          TRACEBACK; status=1; return
       end if

       deallocate(glbsz)

    end if

  END SUBROUTINE CHECK_DIST_ARR
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    UPDATE_HALO_IBAND_1D_R
  !
  ! !DESCRIPTION: update halo cells of a vector distributed along latitudes 
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE UPDATE_HALO_IBAND_1D_R( DistGrid, dist_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in)    :: DistGrid
    integer,         intent(in)    :: halo
    real,            intent(inout) :: dist_array(DistGrid%j_strt-halo:)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out)   :: status
    !
    ! !REVISION HISTORY: 
    !    7 Jan 2016 - Ph. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'update_halo_iband_1d_r'
    integer :: j0, j1

#ifdef MPI

    integer :: stat(MPI_STATUS_SIZE,4), req(4)
    integer :: k, sz(1), tag_snd(2), tag_rcv(2)
    integer :: ijsr(2,2), nghbr(2)

    ! check input
    if ( halo == 0 ) return
    sz = shape(dist_array)
    j0 = DistGrid%j_strt        
    j1 = DistGrid%j_stop    

    ! degenerate case
    if (npe_lat==1) return

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, iband=.true. )
       IF_NOTOK_RETURN(status=1)
    end if

    ! Buffers anchors
    ! ----------------    
    ijsr(:,1) = (/ j1-halo+1,      j0 /) ! j-start location of north and south buffers to send
    ijsr(:,2) = (/      j1+1, j0-halo /) ! j-start location of north and south buffers to receive

    ! Communicate
    ! ---------------- ! only south and north
    tag_snd = (/2,4/)
    tag_rcv = (/4,2/)
    nghbr   = (/2,4/)

    neigh : do k=1,2 

       call MPI_ISEND( dist_array( ijsr(k,1)), halo, my_real, DistGrid%neighbors(nghbr(k)), tag_snd(k), localComm, req(k), ierr)
       call MPI_IRECV( dist_array( ijsr(k,2)), halo, my_real, DistGrid%neighbors(nghbr(k)), tag_rcv(k), localComm, req(k+2), ierr)

    end do neigh

    call MPI_WAITALL(4, req, stat, ierr)
    IF_NOTOK_MPI(status=1)

#endif

    status = 0

  END SUBROUTINE UPDATE_HALO_IBAND_1D_R
  !EOC
    
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    UPDATE_HALO_JBAN_1D_R
  !
  ! !DESCRIPTION: update halo cells of a decomposed zonal vector
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE UPDATE_HALO_JBAND_1D_R( DistGrid, dist_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in)    :: DistGrid
    integer,         intent(in)    :: halo
    real,            intent(inout) :: dist_array(DistGrid%i_strt-halo:)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out)   :: status
    !
    ! !REVISION HISTORY: 
    !   20 Feb 2012 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'update_halo_jband_1d_r'
    integer :: i0, i1
    
#ifdef MPI

    integer :: stat(MPI_STATUS_SIZE,4), req(4)
    integer :: k, sz(1), tag_snd(2), tag_rcv(2)
    integer :: ijsr(2,2), nghbr(2)
    
    ! check input
    if ( halo == 0 ) return
    sz = shape(dist_array)
    i0 = DistGrid%i_strt        
    i1 = DistGrid%i_stop    

    ! degenerate case
    if (npe_lon==1) then
       if (DistGrid%is_periodic) then
          dist_array(i0-halo:i0-1) = dist_array(i1-halo+1:i1)
          dist_array(i1+1:i1+halo) = dist_array(i0:i0+halo-1)
       end if
       return
    end if

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, jband=.true. )
       IF_NOTOK_RETURN(status=1)
    end if


    ! Buffers anchors
    ! ----------------
    ijsr(:,1) = (/      i0, i1+1-halo /) ! i-start location of buffer to send
    ijsr(:,2) = (/ i0-halo,      i1+1 /) ! i-start location of buffer to receive

    
    ! Communicate
    ! ---------------- ! only east and west
    tag_snd = (/1,3/)
    tag_rcv = (/3,1/)
    nghbr   = (/1,3/)
    
    neigh : do k=1,2

       call MPI_ISEND( dist_array( ijsr(k,1)), halo, my_real, DistGrid%neighbors(nghbr(k)), tag_snd(k), localComm, req(k), ierr)
       call MPI_IRECV( dist_array( ijsr(k,2)), halo, my_real, DistGrid%neighbors(nghbr(k)), tag_rcv(k), localComm, req(k+2), ierr)

    end do neigh

    call MPI_WAITALL(4, req, stat, ierr)
    IF_NOTOK_MPI(status=1)

#else
    
    if ((halo/=0).and.DistGrid%is_periodic) then

       i0  = DistGrid%i_strt        
       i1  = DistGrid%i_stop    

       dist_array(i0-halo:i0-1) = dist_array(i1-halo+1:i1)
       dist_array(i1+1:i1+halo) = dist_array(i0:i0+halo-1)

    end if

#endif

    status = 0

  END SUBROUTINE UPDATE_HALO_JBAND_1D_R
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    UPDATE_HALO_JBAND_2D_R
  !
  ! !DESCRIPTION: update halo cells of a decomposed zonal 2d slice
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE UPDATE_HALO_JBAND_2D_R( DistGrid, dist_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in)    :: DistGrid
    integer,         intent(in)    :: halo
    real,            intent(inout) :: dist_array(DistGrid%i_strt-halo:,:)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out)   :: status
    !
    ! !REVISION HISTORY: 
    !   20 Feb 2012 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'update_halo_jband_2d_r'
    integer :: i0, i1

#ifdef MPI

    integer :: stat(MPI_STATUS_SIZE,4), req(4), wetype
    integer :: k, sz(2), tag_snd(2), tag_rcv(2)
    integer :: ijsr(2,2), nghbr(2)

    status = 0
    
    ! check input
    if ( halo == 0 ) return

    sz = shape(dist_array)
    i0 = DistGrid%i_strt        
    i1 = DistGrid%i_stop    

    ! degenerate case
    if (npe_lon==1) then
       if (DistGrid%is_periodic) then
          dist_array(i0-halo:i0-1,:) = dist_array(i1-halo+1:i1,:)
          dist_array(i1+1:i1+halo,:) = dist_array(i0:i0+halo-1,:)
       end if
       return
    end if

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, jband=.true. )
       IF_NOTOK_RETURN(status=1)
    end if

    ! Buffers anchors
    ! ----------------
    ijsr(:,1) = (/      i0, i1+1-halo /) ! i-start location of buffer to send
    ijsr(:,2) = (/ i0-halo,      i1+1 /) ! i-start location of buffer to receive

    ! pack data
    !----------
    call MPI_TYPE_VECTOR (sz(2), halo, sz(1), my_real, wetype, ierr)
    call MPI_TYPE_COMMIT (wetype, ierr)
    
    ! Communicate
    ! ----------------
    tag_snd = (/1,3/)
    tag_rcv = (/3,1/)
    nghbr   = (/1,3/)

    neigh : do k=1,2 ! only east and west

       call MPI_ISEND( dist_array(ijsr(k,1),1), 1, wetype, DistGrid%neighbors(nghbr(k)), tag_snd(k), localComm, req(k), ierr)
       call MPI_IRECV( dist_array(ijsr(k,2),1), 1, wetype, DistGrid%neighbors(nghbr(k)), tag_rcv(k), localComm, req(k+2), ierr)

    end do neigh

    call MPI_WAITALL(4, req, stat, ierr)
    IF_NOTOK_MPI(status=1)

    call MPI_TYPE_FREE(wetype, ierr)

#else
    
    if ((halo/=0).and.DistGrid%is_periodic) then

       i0  = DistGrid%i_strt        
       i1  = DistGrid%i_stop    

       dist_array(i0-halo:i0-1,:) = dist_array(i1-halo+1:i1,:)
       dist_array(i1+1:i1+halo,:) = dist_array(i0:i0+halo-1,:)

    end if

#endif

    status = 0

  END SUBROUTINE UPDATE_HALO_JBAND_2D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    UPDATE_HALO_JBAND_3D_R
  !
  ! !DESCRIPTION: update halo cells of a decomposed zonal 3d slice
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE UPDATE_HALO_JBAND_3D_R( DistGrid, dist_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in)    :: DistGrid
    integer,         intent(in)    :: halo
    real,            intent(inout) :: dist_array(DistGrid%i_strt-halo:,:,:)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out)   :: status
    !
    ! !REVISION HISTORY: 
    !   20 Feb 2012 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'update_halo_jband_3d_r'
    integer :: i0, i1

#ifdef MPI

    integer :: stat(MPI_STATUS_SIZE,4), req(4), wetype
    integer :: k, sz(3), tag_snd(2), tag_rcv(2)
    integer :: ijsr(2,2), nghbr(2)

    status = 0

    ! check input
    if ( halo == 0 ) return

    sz = shape(dist_array)
    i0 = DistGrid%i_strt        
    i1 = DistGrid%i_stop    

    ! degenerate case
    if (npe_lon==1) then
       if (DistGrid%is_periodic) then
          dist_array(i0-halo:i0-1,:,:) = dist_array(i1-halo+1:i1,:,:)
          dist_array(i1+1:i1+halo,:,:) = dist_array(i0:i0+halo-1,:,:)
       end if
       return
    end if

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, jband=.true. )
       IF_NOTOK_RETURN(status=1)
    end if

    ! Buffers anchors
    ! ----------------
    ijsr(:,1) = (/      i0, i1+1-halo /) ! i-start location of buffer to send
    ijsr(:,2) = (/ i0-halo,      i1+1 /) ! i-start location of buffer to receive

    ! pack data
    !----------
    call MPI_TYPE_VECTOR (sz(2)*sz(3), halo, sz(1), my_real, wetype, ierr)
    call MPI_TYPE_COMMIT (wetype, ierr)

    ! Communicate
    ! ----------------
    tag_snd = (/1,3/)
    tag_rcv = (/3,1/)
    nghbr   = (/1,3/)

    neigh : do k=1,2 ! only east and west

       call MPI_ISEND( dist_array(ijsr(k,1),1,1), 1, wetype, DistGrid%neighbors(nghbr(k)), tag_snd(k), localComm, req(k), ierr)
       call MPI_IRECV( dist_array(ijsr(k,2),1,1), 1, wetype, DistGrid%neighbors(nghbr(k)), tag_rcv(k), localComm, req(k+2), ierr)

    end do neigh

    call MPI_WAITALL(4, req, stat, ierr)
    IF_NOTOK_MPI(status=1)

    call MPI_TYPE_FREE(wetype, ierr)

#else

    if ((halo/=0).and.DistGrid%is_periodic) then

       i0  = DistGrid%i_strt        
       i1  = DistGrid%i_stop    

       dist_array(i0-halo:i0-1,:,:) = dist_array(i1-halo+1:i1,:,:)
       dist_array(i1+1:i1+halo,:,:) = dist_array(i0:i0+halo-1,:,:)

    end if

#endif

    status = 0

  END SUBROUTINE UPDATE_HALO_JBAND_3D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    UPDATE_HALO_JBAND_4D_R
  !
  ! !DESCRIPTION: update halo cells of a decomposed zonal 4d slice
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE UPDATE_HALO_JBAND_4D_R( DistGrid, dist_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in)    :: DistGrid
    integer,         intent(in)    :: halo
    real,            intent(inout) :: dist_array(DistGrid%i_strt-halo:,:,:,:)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out)   :: status
    !
    ! !REVISION HISTORY: 
    !   20 Feb 2012 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'update_halo_jband_4d_r'
    integer :: i0, i1

#ifdef MPI

    integer :: stat(MPI_STATUS_SIZE,4), req(4), wetype
    integer :: k, sz(4), tag_snd(2), tag_rcv(2)
    integer :: ijsr(2,2), nghbr(2)

    status = 0

    ! check input
    if ( halo == 0 ) return

    sz = shape(dist_array)
    i0 = DistGrid%i_strt        
    i1 = DistGrid%i_stop    

    ! degenerate case
    if (npe_lon==1) then
       if (DistGrid%is_periodic) then
          dist_array(i0-halo:i0-1,:,:,:) = dist_array(i1-halo+1:i1,:,:,:)
          dist_array(i1+1:i1+halo,:,:,:) = dist_array(i0:i0+halo-1,:,:,:)
       end if
       return
    end if

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, jband=.true. )
       IF_NOTOK_RETURN(status=1)
    end if

    ! Buffers anchors
    ! ----------------
    ijsr(:,1) = (/      i0, i1+1-halo /) ! i-start location of buffer to send
    ijsr(:,2) = (/ i0-halo,      i1+1 /) ! i-start location of buffer to receive

    ! pack data
    !----------
    call MPI_TYPE_VECTOR (sz(2)*sz(3)*sz(4), halo, sz(1), my_real, wetype, ierr)
    call MPI_TYPE_COMMIT (wetype, ierr)

    ! Communicate
    ! ----------------
    tag_snd = (/1,3/)
    tag_rcv = (/3,1/)
    nghbr   = (/1,3/)

    neigh : do k=1,2 ! only east and west

       call MPI_ISEND( dist_array(ijsr(k,1),1,1,1), 1, wetype, DistGrid%neighbors(nghbr(k)), tag_snd(k), localComm, req(k), ierr)
       call MPI_IRECV( dist_array(ijsr(k,2),1,1,1), 1, wetype, DistGrid%neighbors(nghbr(k)), tag_rcv(k), localComm, req(k+2), ierr)

    end do neigh

    call MPI_WAITALL(4, req, stat, ierr)
    IF_NOTOK_MPI(status=1)

    call MPI_TYPE_FREE(wetype, ierr)

#else

    if ((halo/=0).and.DistGrid%is_periodic) then

       i0  = DistGrid%i_strt        
       i1  = DistGrid%i_stop    

       dist_array(i0-halo:i0-1,:,:,:) = dist_array(i1-halo+1:i1,:,:,:)
       dist_array(i1+1:i1+halo,:,:,:) = dist_array(i0:i0+halo-1,:,:,:)

    end if

#endif

    status = 0

  END SUBROUTINE UPDATE_HALO_JBAND_4D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    UPDATE_HALO_2D_R
  !
  ! !DESCRIPTION: update halo cells of a distributed 2D real array
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE UPDATE_HALO_2D_R( DistGrid, dist_array, halo, status, i_only, j_only )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in)    :: DistGrid
    integer,         intent(in)    :: halo
    real,            intent(inout) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:)
    logical, optional, intent(in)  :: i_only ! update East & West halo only
    logical, optional, intent(in)  :: j_only ! update North & South halo only
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out)   :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'update_halo_2d_r'
    integer :: i1, i2, j1, j2

#ifdef MPI
    
    integer :: stat(MPI_STATUS_SIZE,8), req(8)
    integer :: k, sz(2), tag_snd(4), tag_rcv(4)
    integer :: srtype(4), ijsr(4,4)
    
    ! check input
    if ( halo == 0 ) return
    sz = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status )
       IF_NOTOK_RETURN(status=1)
    end if
    
    ! get types and I/J start 
    CALL GET_HALO_TYPE( DistGrid, sz, halo, my_real, srtype, ijsr, status )
    IF_NOTOK_RETURN(status=1)

    ! Communicate
    tag_snd = (/1,2,3,4/)
    tag_rcv = (/3,4,1,2/)
    
    neigh : do k=1,4
       
       call MPI_ISEND( dist_array( ijsr(k,1), ijsr(k,2)), 1, srtype(k), DistGrid%neighbors(k), tag_snd(k), localComm, req(k), ierr)
       call MPI_IRECV( dist_array( ijsr(k,3), ijsr(k,4)), 1, srtype(k), DistGrid%neighbors(k), &
            tag_rcv(k), localComm, req(k+4), ierr)

    end do neigh

    call MPI_WAITALL(8, req, stat, ierr)
    IF_NOTOK_MPI(status=1)

    call FREE_DERIVED_TYPE( srtype )

#else

    if ((halo/=0).and.DistGrid%is_periodic) then

       i1 = DistGrid%i_strt
       i2 = DistGrid%i_stop
       j1 = DistGrid%j_strt
       j2 = DistGrid%j_stop
       
       dist_array(i1-halo:i1-1,j1:j2) = dist_array(i2-halo+1:i2,j1:j2)
       dist_array(i2+1:i2+halo,j1:j2) = dist_array(i1:i1+halo-1,j1:j2)

    end if
#endif

    status = 0
    
  END SUBROUTINE UPDATE_HALO_2D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    UPDATE_HALO_2D_I
  !
  ! !DESCRIPTION: update halo cells of a distributed 2D integer array
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE UPDATE_HALO_2D_I( DistGrid, dist_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in)    :: DistGrid
    integer,         intent(in)    :: halo
    integer,         intent(inout) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS:
    ! (1) not tested yet, but the version for 'real' has been...
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'update_halo_2D_i'
    integer :: i1, i2, j1, j2

#ifdef MPI
    
    integer :: stat(MPI_STATUS_SIZE,8), req(8), k, sz(2)
    integer :: srtype(4), ijsr(4,4), tag_snd(4), tag_rcv(4)

    ! check input
    if ( halo == 0 ) return
    sz = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status )
       IF_NOTOK_RETURN(status=1)
    end if

    ! get types and I/J start 
    CALL GET_HALO_TYPE( distGrid, sz, halo, MPI_INTEGER, srtype, ijsr, status )
    IF_NOTOK_RETURN(status=1)

    ! Communicate
    tag_snd = (/1,2,3,4/)
    tag_rcv = (/3,4,1,2/)

    neigh : do k=1,4

       call MPI_ISEND( dist_array( ijsr(k,1), ijsr(k,2)), 1, srtype(k), DistGrid%neighbors(k), tag_snd(k), localComm, req(k), ierr)
       call MPI_IRECV( dist_array( ijsr(k,3), ijsr(k,4)), 1, srtype(k), DistGrid%neighbors(k), &
            tag_rcv(k), localComm, req(k+4), ierr)

    end do neigh

    call MPI_WAITALL(8, req, stat, ierr)
    call FREE_DERIVED_TYPE( srtype )

#else
    
    if ((halo/=0).and.DistGrid%is_periodic) then

       i1 = DistGrid%i_strt
       i2 = DistGrid%i_stop
       j1 = DistGrid%j_strt
       j2 = DistGrid%j_stop

       dist_array(i1-halo:i1-1,j1:j2) = dist_array(i2-halo+1:i2,j1:j2)
       dist_array(i2+1:i2+halo,j1:j2) = dist_array(i1:i1+halo-1,j1:j2)

    end if
#endif

    status = 0

  END SUBROUTINE UPDATE_HALO_2D_I
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    UPDATE_HALO_3D_R
  !
  ! !DESCRIPTION: update halo cells of a distributed 3D real array
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE UPDATE_HALO_3D_R( DistGrid, dist_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in)    :: DistGrid
    integer,         intent(in)    :: halo
    real,            intent(inout) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:,:)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out)   :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'update_halo_3d_r'
    integer :: i1, i2, j1, j2

#ifdef MPI
    
    integer :: stat(MPI_STATUS_SIZE,8), req(8), k, sz(3)
    integer :: srtype(4), ijsr(4,4), tag_snd(4), tag_rcv(4)

    ! check input
    if ( halo == 0 ) return
    sz = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status )
       IF_NOTOK_RETURN(status=1)
    end if

    ! get types and I/J start 
    CALL GET_HALO_TYPE( DistGrid, sz, halo, my_real, srtype, ijsr, status )
    IF_NOTOK_RETURN(status=1)

    ! Communicate
    tag_snd = (/1,2,3,4/)
    tag_rcv = (/3,4,1,2/)

    neigh : do k=1,4

       call MPI_ISEND( dist_array( ijsr(k,1), ijsr(k,2), 1), 1, srtype(k), DistGrid%neighbors(k), tag_snd(k), localComm, &
            req(k), ierr)
       call MPI_IRECV( dist_array( ijsr(k,3), ijsr(k,4), 1), 1, srtype(k), DistGrid%neighbors(k), tag_rcv(k), localComm, &
            req(k+4), ierr)

    end do neigh
    
    call MPI_WAITALL(8, req, stat, ierr)
    IF_NOTOK_MPI(status=1)

    call FREE_DERIVED_TYPE( srtype )

#else
    
    if ((halo/=0).and.DistGrid%is_periodic) then

       i1 = DistGrid%i_strt
       i2 = DistGrid%i_stop
       j1 = DistGrid%j_strt
       j2 = DistGrid%j_stop

       dist_array(i1-halo:i1-1,j1:j2,:) = dist_array(i2-halo+1:i2,j1:j2,:)
       dist_array(i2+1:i2+halo,j1:j2,:) = dist_array(i1:i1+halo-1,j1:j2,:)

    end if
#endif

    status = 0

  END SUBROUTINE UPDATE_HALO_3D_R
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    UPDATE_HALO_4D_R
  !
  ! !DESCRIPTION: update halo cells of a distributed 4D real array
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE UPDATE_HALO_4D_R( DistGrid, dist_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in)    :: DistGrid
    integer,         intent(in)    :: halo
    real,            intent(inout) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:,:,:)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out)   :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'update_halo_4D_r'
    integer :: i1, i2, j1, j2

#ifdef MPI
    
    integer :: stat(MPI_STATUS_SIZE,8), req(8), k, sz(4)
    integer :: srtype(4), ijsr(4,4), tag_snd(4), tag_rcv(4)

    ! check input
    if ( halo == 0 ) return
    sz = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status )
       IF_NOTOK_RETURN(status=1)
    end if

    ! get types and I/J start 
    CALL GET_HALO_TYPE( DistGrid, sz, halo, my_real, srtype, ijsr, status )
    IF_NOTOK_RETURN(status=1)

    ! Communicate
    tag_snd = (/1,2,3,4/)
    tag_rcv = (/3,4,1,2/)

    neigh : do k=1,4

       call MPI_ISEND( dist_array( ijsr(k,1), ijsr(k,2), 1, 1), 1, srtype(k), DistGrid%neighbors(k), tag_snd(k), &
            localComm, req(k), ierr)
       call MPI_IRECV( dist_array( ijsr(k,3), ijsr(k,4), 1, 1), 1, srtype(k), DistGrid%neighbors(k), tag_rcv(k), &
            localComm, req(k+4), ierr)

    end do neigh

    call MPI_WAITALL(8, req, stat, ierr)
    IF_NOTOK_MPI(status=1)
    
    call FREE_DERIVED_TYPE( srtype )

#else
    
    if ((halo/=0).and.DistGrid%is_periodic) then

       i1 = DistGrid%i_strt
       i2 = DistGrid%i_stop
       j1 = DistGrid%j_strt
       j2 = DistGrid%j_stop

       dist_array(i1-halo:i1-1,j1:j2,:,:) = dist_array(i2-halo+1:i2,j1:j2,:,:)
       dist_array(i2+1:i2+halo,j1:j2,:,:) = dist_array(i1:i1+halo-1,j1:j2,:,:)

    end if
#endif

    status = 0

  END SUBROUTINE UPDATE_HALO_4D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GATHER_2D_R
  !
  ! !DESCRIPTION: gather local 2D arrays into a global 2D array  
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GATHER_2D_R( DistGrid, dist_array, glbl_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    integer,         intent(in) :: halo
    real,            intent(in) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:)
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: glbl_array(:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS:
    ! (1) I have not use mpi_gatherv because of varying *receiving* size
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'gather_2d_r'

#ifndef MPI
    glbl_array = dist_array( DistGrid%i_strt:DistGrid%i_stop, DistGrid%j_strt:DistGrid%j_stop )
    status = 0
#else    

    integer :: stat(MPI_STATUS_SIZE), linterior, ginterior(npes-1)
    integer :: i, j, klm, i0, j0, i1, j1, sz(2)

    status=0
    
    ! check input, get derived types
    !-------------------------------

    sz = shape(dist_array)
    
    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, shape(glbl_array))
       IF_NOTOK_RETURN(status=1)
    end if

    CALL GET_INTERIOR_TYPE( DistGrid, sz, my_real, linterior, ginterior, status )
    IF_NOTOK_RETURN(status=1)
    
    i0  = DistGrid%i_strt
    j0  = DistGrid%j_strt        

    ! ------- GATHER -------------
    
    if ( isRoot ) then

       ! get first chunk locally
       i1 = DistGrid%i_stop
       j1 = DistGrid%j_stop

       glbl_array(i0:i1,j0:j1) = dist_array(i0:i1,j0:j1)

       ! get remaining chunks from other pes
       do klm=1,npes-1

          i = DistGrid%bounds(1,klm)
          j = DistGrid%bounds(3,klm)

          call MPI_RECV( glbl_array(i,j), 1, ginterior(klm), klm, 1, &
                         localComm, stat, ierr)

       end do

       call FREE_DERIVED_TYPE( ginterior )

    else

       call MPI_SEND( dist_array(i0,j0), 1, linterior, root, 1, localComm, ierr) 
       call MPI_TYPE_FREE(linterior, ierr)

    end if
#endif
    
  END SUBROUTINE GATHER_2D_R
  !EOC
      
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GATHER_2D_I
  !
  ! !DESCRIPTION: gather local 2D arrays into a global 2D array (integer version)  
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GATHER_2D_I( DistGrid, dist_array, glbl_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    integer,         intent(in) :: halo
    integer,         intent(in) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer, intent(out) :: glbl_array(:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'gather_2d_i'

#ifndef MPI
    glbl_array = dist_array( DistGrid%i_strt:DistGrid%i_stop, DistGrid%j_strt:DistGrid%j_stop )
    status = 0
#else    

    integer :: stat(MPI_STATUS_SIZE), linterior, ginterior(npes-1)
    integer :: i, j, klm, i0, j0, i1, j1, sz(2)

    status=0

    ! check input, get derived types
    !-------------------------------

    sz = shape(dist_array)
    
    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, shape(glbl_array))
       IF_NOTOK_RETURN(status=1)
    end if

    CALL GET_INTERIOR_TYPE( DistGrid, sz, MPI_INTEGER, linterior, ginterior, status )
    IF_NOTOK_RETURN(status=1)
    
    i0  = DistGrid%i_strt
    j0  = DistGrid%j_strt        

    ! ------- GATHER -------------

    if ( isRoot ) then

       ! get first chunk locally
       i1 = DistGrid%i_stop
       j1 = DistGrid%j_stop

       glbl_array(i0:i1,j0:j1) = dist_array(i0:i1,j0:j1)

       ! get remaining chunks from other pes
       do klm=1,npes-1

          i = DistGrid%bounds(1,klm)
          j = DistGrid%bounds(3,klm)

          call MPI_RECV( glbl_array(i,j), 1, ginterior(klm), klm, 1, &
               localComm, stat, ierr)

       end do

       call FREE_DERIVED_TYPE( ginterior )

    else

       call MPI_SEND( dist_array(i0,j0), 1, linterior, root, 1, localComm, ierr) 
       call MPI_TYPE_FREE(linterior, ierr)

    end if
#endif

  END SUBROUTINE GATHER_2D_I
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GATHER_3D_R
  !
  ! !DESCRIPTION: gather local 3D arrays into a global 3D array  
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GATHER_3D_R( DistGrid, dist_array, glbl_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    integer,         intent(in) :: halo
    real,            intent(in) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:,:)
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: glbl_array(:,:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'gather_3d_r'
    
#ifndef MPI
    glbl_array = dist_array( DistGrid%i_strt:DistGrid%i_stop, DistGrid%j_strt:DistGrid%j_stop, : )
    status = 0
#else
    
    integer :: stat(MPI_STATUS_SIZE), linterior, ginterior(npes-1)
    integer :: i, j, klm, i0, j0, i1, j1, sz(3)

    status=0

    ! ------- Check input & get derived types 

    sz  = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, shape(glbl_array))
       IF_NOTOK_RETURN(status=1)
    end if

    CALL GET_INTERIOR_TYPE( DistGrid, sz, my_real, linterior, ginterior, status )
    IF_NOTOK_RETURN(status=1)
    
    i0  = DistGrid%i_strt
    j0  = DistGrid%j_strt        

    ! ------- GATHER -------------

    if ( isRoot ) then

       ! get first chunk locally
       i1 = DistGrid%i_stop
       j1 = DistGrid%j_stop

       glbl_array(i0:i1,j0:j1,:) = dist_array(i0:i1,j0:j1,:)

       ! get remaining chunks from other pes
       do klm=1,npes-1

          i = DistGrid%bounds(1,klm)
          j = DistGrid%bounds(3,klm)

          call MPI_RECV( glbl_array(i,j,1), 1, ginterior(klm), klm, 1, &
                         localComm, stat, ierr)

       end do

       call FREE_DERIVED_TYPE( ginterior )

    else

       call MPI_SEND( dist_array(i0,j0,1), 1, linterior, root, 1, localComm, ierr) 
       call MPI_TYPE_FREE(linterior, ierr)

    end if
#endif
  END SUBROUTINE GATHER_3D_R
  !EOC

  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GATHER_4D_R
  !
  ! !DESCRIPTION: gather local 4D arrays into a global 4D array  
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GATHER_4D_R( DistGrid, dist_array, glbl_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    integer,         intent(in) :: halo
    real,            intent(in) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:,:,:)
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: glbl_array(:,:,:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS:
    ! (1) global array has to really be global on root only
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'gather_4d_r'

#ifndef MPI
    glbl_array = dist_array( DistGrid%i_strt:DistGrid%i_stop, DistGrid%j_strt:DistGrid%j_stop, :, :)
    status = 0
#else    

    integer :: stat(MPI_STATUS_SIZE), linterior, ginterior(npes-1)
    integer :: i, j, klm, i0, j0, i1, j1, sz(4)

    status=0

    ! ------- Check input & get derived types 

    sz  = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, shape(glbl_array))
       IF_NOTOK_RETURN(status=1)
    end if

    CALL GET_INTERIOR_TYPE( DistGrid, sz, my_real, linterior, ginterior, status )
    IF_NOTOK_RETURN(status=1)
    
    i0  = DistGrid%i_strt
    j0  = DistGrid%j_strt        

    ! ------- GATHER -------------

    if ( isRoot ) then  ! RECV

       ! get first chunk locally
       i1 = DistGrid%i_stop
       j1 = DistGrid%j_stop

       glbl_array(i0:i1,j0:j1,:,:) = dist_array(i0:i1,j0:j1,:,:)

       ! send remaining chunks to other pes
       do klm=1,npes-1

          i = DistGrid%bounds(1,klm)
          j = DistGrid%bounds(3,klm)

          call MPI_RECV( glbl_array(i,j,1,1), 1, ginterior(klm), klm, 1, &
                         localComm, stat, ierr)
       end do

       call FREE_DERIVED_TYPE( ginterior )

    else !SEND

       call MPI_SEND( dist_array(i0,j0,1,1), 1, linterior, root, 1, localComm, ierr) 
       call MPI_TYPE_FREE(linterior, ierr)

    end if
#endif
  END SUBROUTINE GATHER_4D_R
  !EOC

  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GATHER_5D_R
  !
  ! !DESCRIPTION: gather local 5D arrays into a global 5D array  
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GATHER_5D_R( DistGrid, dist_array, glbl_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    integer,         intent(in) :: halo
    real,            intent(in) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:,:,:,:)
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: glbl_array(:,:,:,:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS:
    ! (1) global array has to really be global on root only
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'gather_5d_r'

#ifndef MPI
    glbl_array = dist_array( DistGrid%i_strt:DistGrid%i_stop, DistGrid%j_strt:DistGrid%j_stop, :,:,:)
    status = 0
#else    

    integer :: stat(MPI_STATUS_SIZE), linterior, ginterior(npes-1)
    integer :: i, j, klm, i0, j0, i1, j1, sz(5)

    status=0

    ! ------- Check input & get derived types 

    sz  = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, shape(glbl_array))
       IF_NOTOK_RETURN(status=1)
    end if

    CALL GET_INTERIOR_TYPE( DistGrid, sz, my_real, linterior, ginterior, status )
    IF_NOTOK_RETURN(status=1)

    i0  = DistGrid%i_strt
    j0  = DistGrid%j_strt        

    ! ------- GATHER -------------

    if ( isRoot ) then  ! RECV

       ! get first chunk locally
       i1 = DistGrid%i_stop
       j1 = DistGrid%j_stop

       glbl_array(i0:i1,j0:j1,:,:,:) = dist_array(i0:i1,j0:j1,:,:,:)

       ! send remaining chunks to other pes
       do klm=1,npes-1

          i = DistGrid%bounds(1,klm)
          j = DistGrid%bounds(3,klm)

          call MPI_RECV( glbl_array(i,j,1,1,1), 1, ginterior(klm), klm, 1, &
               localComm, stat, ierr)
       end do

       call FREE_DERIVED_TYPE( ginterior )

    else !SEND

       call MPI_SEND( dist_array(i0,j0,1,1,1), 1, linterior, root, 1, localComm, ierr) 
       call MPI_TYPE_FREE(linterior, ierr)

    end if
#endif
  END SUBROUTINE GATHER_5D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    SCATTER_2D_R
  !
  ! !DESCRIPTION: scatter a 2D global real array  
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE SCATTER_2D_R( DistGrid, dist_array, glbl_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    real,            intent(in) :: glbl_array(:,:)
    integer,         intent(in) :: halo
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !   21 Jun 2013 - P. Le Sager - MPI_SEND -> MPI_SSEND to avoid buffering
    !
    ! !REMARKS: exactly the same as GATHER_2D_R, but inverting send/recv
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'scatter_2d_r'

#ifndef MPI
    dist_array( DistGrid%i_strt:DistGrid%i_stop, DistGrid%j_strt:DistGrid%j_stop ) = glbl_array
    status = 0
#else
    
    integer :: stat(MPI_STATUS_SIZE), linterior, ginterior(npes-1)
    integer :: i, j, klm, i0, j0, i1, j1, sz(2)

    status=0
 
    ! ------- Check input & get derived types 

    sz = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, shape(glbl_array))
       IF_NOTOK_RETURN(status=1)
    end if

    CALL GET_INTERIOR_TYPE( DistGrid, sz, my_real, linterior, ginterior, status )
    IF_NOTOK_RETURN(status=1)

    i0  = DistGrid%i_strt
    j0  = DistGrid%j_strt


    ! ------- SCATTER -------------

    if ( isRoot ) then  ! send

       ! get first chunk locally
       i1 = DistGrid%i_stop
       j1 = DistGrid%j_stop

       dist_array(i0:i1,j0:j1) = glbl_array(i0:i1,j0:j1)

       ! send remaining chunks to other pes
       do klm=1,npes-1

          i = DistGrid%bounds(1,klm)
          j = DistGrid%bounds(3,klm)

          call MPI_SSEND( glbl_array(i,j), 1, ginterior(klm), klm, klm, localComm, ierr)
          IF_NOTOK_MPI(status=1)

       end do

       call FREE_DERIVED_TYPE( ginterior )

    else

       call MPI_COMM_RANK(localComm, klm, ierr)
       IF_NOTOK_MPI(status=1)

       call MPI_RECV( dist_array(i0,j0), 1, linterior, root, klm, localComm, stat, ierr)
       IF_NOTOK_MPI(status=1)
       
       call MPI_TYPE_FREE(linterior, ierr)
       IF_NOTOK_MPI(status=1)

    end if
#endif
  END SUBROUTINE SCATTER_2D_R
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    SCATTER_3D_R
  !
  ! !DESCRIPTION: scatter 3D global real array
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE SCATTER_3D_R( DistGrid, dist_array, glbl_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    integer,         intent(in) :: halo
    real,            intent(in) :: glbl_array(:,:,:)
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !   21 Jun 2013 - P. Le Sager - MPI_SEND -> MPI_SSEND to avoid buffering
    !
    ! !REMARKS: global array has to really be global on root only
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'scatter_3d_r'

#ifndef MPI
    dist_array( DistGrid%i_strt:DistGrid%i_stop, DistGrid%j_strt:DistGrid%j_stop, :) = glbl_array
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), linterior, ginterior(npes-1)
    integer :: i, j, klm, i0, j0, i1, j1, sz(3)

    status=0

    ! ------- Check input & get derived types 

    sz  = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, shape(glbl_array))
       IF_NOTOK_RETURN(status=1)
    end if

    CALL GET_INTERIOR_TYPE( DistGrid, sz, my_real, linterior, ginterior, status )
    IF_NOTOK_RETURN(status=1)
    
    i0  = DistGrid%i_strt
    j0  = DistGrid%j_strt        

    ! ------- SCATTER -------------

    if ( isRoot ) then  ! send

       ! get first chunk locally
       i1 = DistGrid%i_stop
       j1 = DistGrid%j_stop

       dist_array(i0:i1,j0:j1,:) = glbl_array(i0:i1,j0:j1,:)

       ! send remaining chunks to other pes
       do klm=1,npes-1

          i = DistGrid%bounds(1,klm)
          j = DistGrid%bounds(3,klm)

          call MPI_SSEND( glbl_array(i,j,1), 1, ginterior(klm), klm, 1, localComm, ierr)
       end do

       call FREE_DERIVED_TYPE( ginterior )

    else

       call MPI_RECV( dist_array(i0,j0,1), 1, linterior, root, 1, localComm, stat, ierr) 
       call MPI_TYPE_FREE(linterior, ierr)

    end if
#endif
  END SUBROUTINE SCATTER_3D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    SCATTER_4D_R
  !
  ! !DESCRIPTION: scatter 4D real array
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE SCATTER_4D_R( DistGrid, dist_array, glbl_array, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    integer,         intent(in) :: halo
    real,            intent(in) :: glbl_array(:,:,:,:)
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:,:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !   21 Jun 2013 - P. Le Sager - MPI_SEND -> MPI_SSEND to avoid buffering
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'scatter_4d_r'

#ifndef MPI
    dist_array( DistGrid%i_strt:DistGrid%i_stop, DistGrid%j_strt:DistGrid%j_stop,:,:) = glbl_array
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), linterior, ginterior(npes-1)
    integer :: i, j, klm, i0, j0, i1, j1, sz(4)

    status=0
    
    ! ------- Check input & get derived types 

    sz  = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, halo, status, shape(glbl_array))
       IF_NOTOK_RETURN(status=1)
    end if

    CALL GET_INTERIOR_TYPE( DistGrid, sz, my_real, linterior, ginterior, status )
    IF_NOTOK_RETURN(status=1)

    i0  = DistGrid%i_strt
    j0  = DistGrid%j_strt        

    ! ------- SCATTER -------------

    if ( isRoot ) then  ! send

       ! get first chunk locally
       i1 = DistGrid%i_stop
       j1 = DistGrid%j_stop
       
       dist_array(i0:i1,j0:j1,:,:) = glbl_array(i0:i1,j0:j1,:,:)

       ! send remaining chunks to other pes
       do klm=1,npes-1

          i = DistGrid%bounds(1,klm)
          j = DistGrid%bounds(3,klm)

          call MPI_SSEND( glbl_array(i,j,1,1), 1, ginterior(klm), klm, 1, localComm, ierr)
       end do

       call FREE_DERIVED_TYPE( ginterior )
       
    else

       call MPI_RECV( dist_array(i0,j0,1,1), 1, linterior, root, 1, localComm, stat, ierr) 
       call MPI_TYPE_FREE(linterior, ierr)

    end if
#endif    
  END SUBROUTINE SCATTER_4D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    SCATTER_IBAND_1D_R
  !
  ! !DESCRIPTION: scatter a meridional real vector (1D) from root
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE SCATTER_IBAND_1D_R( DistGrid, dist_array, glbl_array, status, iref )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: glbl_array(:)
    integer, optional, intent(in) :: iref  ! to find targets, default=1
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: dist_array(DistGrid%j_strt:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !   21 Jun 2013 - P. Le Sager - MPI_SEND -> MPI_SSEND to avoid buffering
    !
    ! !REMARKS: 1D version, along J index, of scatter_2d_r
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'scatter_iband_1d_r'

#ifndef MPI
    dist_array( DistGrid%j_strt:DistGrid%j_stop ) = glbl_array
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE)
    integer :: x_iref, n, klm, i0, j0, i1, j1, sz(1), i0t, j0t, i1t, j1t

    status=0

    ! ------- Check inputs
    x_iref=1
    if(present(iref)) x_iref=iref
    
    sz = shape(dist_array)

    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop
    
    ! ------- SEND/RECV -------------
    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, shape(glbl_array), iband=.true.)
       IF_NOTOK_RETURN(status=1)
    end if

    if ( isRoot ) then

       ! local case
       if((x_iref>=i0).and.(x_iref<=i1)) dist_array(j0:j1) = glbl_array(j0:j1)

       ! send remaining chunks to other pes
       do klm=1,npes-1

          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! is klm a target processor?
          if((x_iref<i0t).or.(x_iref>i1t))cycle

          n=j1t-j0t+1
          call MPI_SSEND( glbl_array(j0t), n, my_real, klm, 1, localComm, ierr)
       end do
       
    else
       
       ! are we on a target processor?       
       if((x_iref<i0).or.(x_iref>i1))return
       
       n=j1-j0+1
       call MPI_RECV( dist_array(j0), n, my_real, root, 1, localComm, stat, ierr) 

    end if
#endif
  END SUBROUTINE SCATTER_IBAND_1D_R
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    SCATTER_JBAND_1D_R
  !
  ! !DESCRIPTION: scatter a zonal real vector (1D) from root
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE SCATTER_JBAND_1D_R( DistGrid, dist_array, glbl_array, status, jref )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: glbl_array(:)
    integer, optional, intent(in) :: jref  ! to find targets, default=1
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: dist_array(DistGrid%i_strt:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !   21 Jun 2013 - P. Le Sager - MPI_SEND -> MPI_SSEND to avoid buffering
    !
    ! !REMARKS: 1D version, along I index, of scatter_2d_r
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'scatter_jband_1d_r'

#ifndef MPI
    dist_array( DistGrid%i_strt:DistGrid%i_stop ) = glbl_array
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE)
    integer :: x_jref, n, klm, i0, j0, i1, j1, sz(1), i0t, j0t, i1t, j1t

    status=0

    ! ------- Check inputs
    x_jref=1
    if(present(jref)) x_jref=jref

    sz = shape(dist_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, shape(glbl_array), jband=.true.)
       IF_NOTOK_RETURN(status=1)
    end if

    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop

    ! ------- SEND/RECV -------------

    if ( isRoot ) then

       ! local case
       if((x_jref>=j0).and.(x_jref<=j1)) dist_array(i0:i1) = glbl_array(i0:i1)

       ! send remaining chunks to other pes
       do klm=1,npes-1

          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! is klm a target processor?
          if((x_jref<j0t).or.(x_jref>j1t))cycle
          n=i1t-i0t+1
          call MPI_SSEND( glbl_array(i0t), n, my_real, klm, 1, localComm, ierr)
       end do

    else

       ! are we on a target processor?       
       if((x_jref<j0).or.(x_jref>j1))return

       n=i1-i0+1
       call MPI_RECV( dist_array(i0), n, my_real, root, 1, localComm, stat, ierr) 

    end if
#endif
  END SUBROUTINE SCATTER_JBAND_1D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    SCATTER_JBAND_2D_R
  !
  ! !DESCRIPTION: scatter a zonal slice (2D) from root
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE SCATTER_JBAND_2D_R( DistGrid, dist_array, glbl_array, status, jref )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: glbl_array(:,:)
    integer, optional, intent(in) :: jref  ! to find targets, default=1
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: dist_array(DistGrid%i_strt:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !   21 Jun 2013 - P. Le Sager - MPI_SEND -> MPI_SSEND to avoid buffering
    !
    ! !REMARKS: 2D version, along I index, of scatter_3d_r
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'scatter_jband_2d_r'

#ifndef MPI
    dist_array( DistGrid%i_strt:DistGrid%i_stop,: ) = glbl_array
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), subarr
    integer :: i0, j0, i1, j1, i0t, j0t, i1t, j1t
    integer :: x_jref, n, klm, sz(2), gsz(2)

    status=0

    ! ------- Check inputs
    x_jref=1
    if(present(jref)) x_jref=jref

    sz = shape(dist_array)
    gsz = shape(glbl_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, gsz, jband=.true.)
       IF_NOTOK_RETURN(status=1)
    end if

    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop

    ! ------- SEND/RECV -------------

    if ( isRoot ) then

       !local case
       if((x_jref>=j0).and.(x_jref<=j1)) dist_array(i0:i1,:) = glbl_array(i0:i1,:)

       ! send remaining chunks to other pes
       do klm=1,npes-1

          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! if klm is a target processor, pack and send
          if((x_jref<j0t).or.(x_jref>j1t))cycle
          n=i1t-i0t+1
          call MPI_TYPE_VECTOR (sz(2), n, gsz(1), my_real, subarr, ierr)
          call MPI_TYPE_COMMIT (subarr, ierr)
          call MPI_SSEND( glbl_array(i0t,1), 1, subarr, klm, 1, localComm, ierr)
          call MPI_TYPE_FREE (subarr, ierr)
       end do

    else

       ! are we on a target processor?       
       if((x_jref<j0).or.(x_jref>j1))return

       n=i1-i0+1
       call MPI_RECV( dist_array(i0,1), n*sz(2), my_real, root, 1, localComm, stat, ierr) 

    end if
#endif
  END SUBROUTINE SCATTER_JBAND_2D_R
  !EOC

  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    SCATTER_IBAND_2D_R
  !
  ! !DESCRIPTION: scatter a meridional real array (2D) from root
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE SCATTER_IBAND_2D_R( DistGrid, dist_array, glbl_array, status, iref )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: glbl_array(:,:)
    integer, optional, intent(in) :: iref  ! to find targets, default=1
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: dist_array(DistGrid%j_strt:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !   21 Jun 2013 - P. Le Sager - MPI_SEND -> MPI_SSEND to avoid buffering
    !
    ! !REMARKS: 2D version, along J index, of scatter_3d_r
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'scatter_iband_2d_r'

#ifndef MPI
    dist_array( DistGrid%j_strt:DistGrid%j_stop,: ) = glbl_array
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), subarr
    integer :: i0, j0, i1, j1, i0t, j0t, i1t, j1t
    integer :: x_iref, n, klm, sz(2), gsz(2)

    status=0

    ! ------- Check inputs
    x_iref=1
    if(present(iref)) x_iref=iref

    sz = shape(dist_array)
    gsz = shape(glbl_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, gsz, iband=.true.)
       IF_NOTOK_RETURN(status=1)
    end if

    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop

    ! ------- SEND/RECV -------------

    if ( isRoot ) then

       ! local case
       if((x_iref>=i0).and.(x_iref<=i1)) dist_array(j0:j1,:) = glbl_array(j0:j1,:)

       ! send remaining chunks to other pes
       do klm=1,npes-1

          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! is klm a target processor?       
          if((x_iref<i0t).or.(x_iref>i1t))cycle
          
          n=j1t-j0t+1
          call MPI_TYPE_VECTOR (sz(2), n, gsz(1), my_real, subarr, ierr)
          call MPI_TYPE_COMMIT (subarr, ierr)
          call MPI_SSEND( glbl_array(j0t,1), 1, subarr, klm, 1, localComm, ierr)
          call MPI_TYPE_FREE (subarr, ierr)
          
       end do

    else

       ! are we on a target processor?       
       if((x_iref<i0).or.(x_iref>i1))return

       n=j1-j0+1
       call MPI_RECV( dist_array(j0,1), n*sz(2), my_real, root, 1, localComm, stat, ierr)

    end if
#endif
  END SUBROUTINE SCATTER_IBAND_2D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    SCATTER_JBAND_3D_R
  !
  ! !DESCRIPTION: scatter a zonal slice (2D) from root
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE SCATTER_JBAND_3D_R( DistGrid, dist_array, glbl_array, status, jref, rowcom )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: glbl_array(:,:,:)
    integer, optional, intent(in) :: jref  ! to find targets, default=1
    logical, optional, intent(in) :: rowcom ! to scatter from row_root instead of global root
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: dist_array(DistGrid%i_strt:,:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !
    ! !REMARKS: 2D version, along I index, of scatter_3d_r
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'scatter_jband_3d_r'

#ifndef MPI
    dist_array( DistGrid%i_strt:DistGrid%i_stop,:,: ) = glbl_array
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), subarr
    integer :: i0, j0, i1, j1, i0t, j0t, i1t, j1t
    integer :: x_jref, n, klm, sz(3), gsz(3), slab, tgt_root
    integer(kind=MPI_ADDRESS_KIND) :: sizeoftype, lb, stride
    logical :: selectRoot

    status=0

    ! ------- Check inputs
    x_jref=1
    if(present(jref)) x_jref=jref

    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop

    ! by default scatter from global root
    selectRoot = isRoot
    tgt_root = root
    if (present(rowcom)) then
       if (rowcom) then
          selectRoot = isRowRoot.and.(x_jref>=j0).and.(x_jref<=j1)
          tgt_root = RowRootId
       endif
    endif
    
    sz = shape(dist_array)
    gsz = shape(glbl_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, gsz, jband=.true., has_global=selectRoot)
       IF_NOTOK_RETURN(status=1)
    end if

    ! ------- SEND/RECV -------------

    if ( selectRoot ) then

       !local case
       if((x_jref>=j0).and.(x_jref<=j1)) dist_array(i0:i1,:,:) = glbl_array(i0:i1,:,:)

       ! send remaining chunks to other pes
       do klm=0,npes-1

          if (klm==myid) cycle  ! skip local case
          
          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! if klm is a target processor, pack and send
          if((x_jref<j0t).or.(x_jref>j1t))cycle
          n=i1t-i0t+1
          call MPI_TYPE_VECTOR (sz(2), n, gsz(1), my_real, slab, ierr)
          CALL MPI_TYPE_GET_EXTENT( my_real, lb, sizeoftype, ierr)
          stride = gsz(1)*gsz(2)*sizeoftype
          call MPI_TYPE_CREATE_HVECTOR (sz(3), 1, stride, slab, subarr, ierr)
          call MPI_TYPE_FREE (slab, ierr)
          call MPI_TYPE_COMMIT (subarr, ierr)
          call MPI_SSEND( glbl_array(i0t,1,1), 1, subarr, klm, 1, localComm, ierr)
          call MPI_TYPE_FREE (subarr, ierr)
       end do

    else

       ! are we on a target processor?       
       if((x_jref<j0).or.(x_jref>j1))return

       n=i1-i0+1
       call MPI_RECV( dist_array(i0,1,1), n*sz(2)*sz(3), my_real, tgt_root, 1, localComm, stat, ierr) 

    end if
#endif
  END SUBROUTINE SCATTER_JBAND_3D_R
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    SCATTER_JBAND_4D_R
  !
  ! !DESCRIPTION: scatter a zonal slice (4D) from root
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE SCATTER_JBAND_4D_R( DistGrid, dist_array, glbl_array, status, jref, rowcom )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: glbl_array(:,:,:,:)
    integer, optional, intent(in) :: jref  ! to find targets, default=1
    logical, optional, intent(in) :: rowcom ! to scatter from row_root instead of global root
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: dist_array(DistGrid%i_strt:,:,:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   17 Feb 2015 - Ph. Le Sager - v0
    !
    ! !REMARKS: 2D version, along I index, of scatter_4d_r
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'scatter_jband_4d_r'

#ifndef MPI
    dist_array( DistGrid%i_strt:DistGrid%i_stop,:,: ) = glbl_array
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), subarr
    integer :: i0, j0, i1, j1, i0t, j0t, i1t, j1t
    integer :: x_jref, n, klm, sz(4), gsz(4), slab, tgt_root
    integer(kind=MPI_ADDRESS_KIND) :: sizeoftype, lb, stride
    logical :: selectRoot

    status=0

    ! ------- Check inputs
    x_jref=1
    if(present(jref)) x_jref=jref

    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop

    ! by default scatter from global root
    selectRoot = isRoot
    tgt_root = root
    if (present(rowcom)) then
       if (rowcom) then
          selectRoot = isRowRoot.and.(x_jref>=j0).and.(x_jref<=j1)
          tgt_root = RowRootId
       endif
    endif

    sz = shape(dist_array)
    gsz = shape(glbl_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, gsz, jband=.true., has_global=selectRoot)
       IF_NOTOK_RETURN(status=1)
    end if

    ! ------- SEND/RECV -------------

    if ( selectRoot ) then

       !local case
       if((x_jref>=j0).and.(x_jref<=j1)) dist_array(i0:i1,:,:,:) = glbl_array(i0:i1,:,:,:)

       ! send remaining chunks to other pes
       do klm=0,npes-1

          if (klm==myid) cycle  ! skip local case

          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! if klm is a target processor, pack and send
          if((x_jref<j0t).or.(x_jref>j1t))cycle
          n=i1t-i0t+1
          call MPI_TYPE_VECTOR (sz(2), n, gsz(1), my_real, slab, ierr)
          CALL MPI_TYPE_GET_EXTENT( my_real, lb, sizeoftype, ierr)
          stride = gsz(1)*gsz(2)*sizeoftype
          call MPI_TYPE_CREATE_HVECTOR (sz(3)*sz(4), 1, stride, slab, subarr, ierr)
          call MPI_TYPE_FREE (slab, ierr)
          call MPI_TYPE_COMMIT (subarr, ierr)
          call MPI_SSEND( glbl_array(i0t,1,1,1), 1, subarr, klm, 1, localComm, ierr)
          call MPI_TYPE_FREE (subarr, ierr)
       end do

    else

       ! are we on a target processor?       
       if((x_jref<j0).or.(x_jref>j1))return

       n=i1-i0+1
       call MPI_RECV( dist_array(i0,1,1,1), n*sz(2)*sz(3)*sz(4), my_real, tgt_root, 1, localComm, stat, ierr) 

    end if
#endif
  END SUBROUTINE SCATTER_JBAND_4D_R
  !EOC
    
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GATHER_JBAND_2D_R
  !
  ! !DESCRIPTION: gather a zonal slice (2D) on root. For 2D arrays, with first
  !               dimension distributed along I (making it a J-band), and the
  !               other dimension is *not* distributed along J. For example: 
  !                 [i1:i2, lev], or [i1:i2, trac]
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GATHER_JBAND_2D_R( DistGrid, dist_array, glbl_array, status, jref )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: dist_array(DistGrid%i_strt:,:)
    integer,           intent(in) :: jref  ! to find sources
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: glbl_array(:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS: 
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'gather_jband_2d_r'

#ifndef MPI
    glbl_array = dist_array( DistGrid%i_strt:DistGrid%i_stop,: )
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), subarr
    integer :: i0, j0, i1, j1, i0t, j0t, i1t, j1t
    integer :: n, klm, sz(2), gsz(2)

    status=0

    ! ------- Check inputs
    sz = shape(dist_array)
    gsz = shape(glbl_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, gsz, jband=.true.)
       IF_NOTOK_RETURN(status=1)
    end if

    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop

    ! ------- SEND/RECV -------------
    if ( isRoot ) then

       ! local case
       if((jref>=j0).and.(jref<=j1)) glbl_array(i0:i1,:) = dist_array(i0:i1,:)

       ! receive remaining chunks from other pes
       do klm=1, npes-1
          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! if klm is a source processor, receive from it
          if((jref<j0t).or.(jref>j1t))cycle
          n=i1t-i0t+1
          call MPI_TYPE_VECTOR (sz(2), n, gsz(1), my_real, subarr, ierr)
          call MPI_TYPE_COMMIT (subarr, ierr)
          call MPI_RECV( glbl_array(i0t,1), 1, subarr, klm, jref, localComm, stat, ierr)
          call MPI_TYPE_FREE (subarr, ierr)
       end do

    else

       ! are we on a src processor?       
       if((jref<j0).or.(jref>j1))return
       n=i1-i0+1
       call MPI_SEND( dist_array(i0,1), n*sz(2), my_real, root, jref, localComm, ierr) 

    end if
#endif
  END SUBROUTINE GATHER_JBAND_2D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GATHER_JBAND_3D_R
  !
  ! !DESCRIPTION: Gather a zonal slab (3D) on root or rowroot(jref) [i.e. the
  !               root of the row of procs].
  !               Local arrays [i1:i2, a:b, c:d] are gathered into the root
  !               proc as [1:im, 1:b-a+1, 1:d-c+1]. Caller has to ensure that at least
  !                 **ALL** the ROW procs call this routine, plus root if needed.
  !               No halo possible yet. 
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GATHER_JBAND_3D_R( DistGrid, dist_array, glbl_array, status, jref, rowcom)
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: dist_array(DistGrid%i_strt:,:,:)
    integer,           intent(in) :: jref   ! to find sources (defines the row we want)
    logical, optional, intent(in) :: rowcom ! to gather on row_root instead of global root
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: glbl_array(:,:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS: use in budget for gathering fluxes, advect_cfl
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'gather_jband_3d_r'

#ifndef MPI
    glbl_array = dist_array( DistGrid%i_strt:DistGrid%i_stop,:,:)
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), subarr
    integer :: i0, j0, i1, j1, i0t, j0t, i1t, j1t
    integer :: n, klm, sz(3), gsz(3), slab, tgt_root
    integer(kind=MPI_ADDRESS_KIND) :: sizeoftype, lb, stride
    logical :: selectRoot

    status=0

    ! ------- Check inputs
     
    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop

    ! by default gather into global root
    selectRoot = isRoot
    tgt_root = root
    if (present(rowcom)) then
       if (rowcom) then
          selectRoot = isRowRoot.and.(jref>=j0).and.(jref<=j1)
          tgt_root = RowRootId
       endif
    endif
    
    sz = shape(dist_array)
    gsz = shape(glbl_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, gsz, jband=.true., has_global=selectRoot)
       IF_NOTOK_RETURN(status=1)
    end if
    
    ! ------- SEND/RECV -------------
    if ( selectRoot ) then

       ! local case
       if((jref>=j0).and.(jref<=j1)) glbl_array(i0:i1,:,:) = dist_array(i0:i1,:,:)

       ! receive remaining chunks from other pes
       do klm=0,npes-1

          if (klm==myid) cycle  ! skip local case
          
          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! if klm is a source processor, receive from it
          if((jref<j0t).or.(jref>j1t))cycle
          n=i1t-i0t+1
          call MPI_TYPE_VECTOR (sz(2), n, gsz(1), my_real, slab, ierr)
          CALL MPI_TYPE_GET_EXTENT( my_real, lb, sizeoftype, ierr)
          stride = gsz(1)*gsz(2)*sizeoftype
          call MPI_TYPE_CREATE_HVECTOR (sz(3), 1, stride, slab, subarr, ierr)
          call MPI_TYPE_FREE (slab, ierr)
          call MPI_TYPE_COMMIT (subarr, ierr)
          call MPI_RECV( glbl_array(i0t,1,1), 1, subarr, klm, jref, localComm, stat, ierr)
          call MPI_TYPE_FREE (subarr, ierr)

       end do

    else

       ! are we on a src processor?       
       if((jref<j0).or.(jref>j1))return
       n=i1-i0+1
       call MPI_SEND( dist_array(i0,1,1), n*sz(2)*sz(3), my_real, tgt_root, jref, localComm, ierr) 

    end if
#endif
  END SUBROUTINE GATHER_JBAND_3D_R
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GATHER_JBAND_4D_R
  !
  ! !DESCRIPTION: Gather a zonal slab (4D) on root or rowroot(jref) [i.e. the
  !               root of the row of procs].
  !               Local arrays [i1:i2, a:b, c:d, e:f] are gathered into the root
  !               proc as [1:im, 1:b-a+1, 1:d-c+1, 1:f-e+1]. Caller has to
  !               ensure that at least **ALL** the ROW procs call this routine,
  !               plus root if needed. 
  !               No halo possible yet. 
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GATHER_JBAND_4D_R( DistGrid, dist_array, glbl_array, status, jref, rowcom)
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: dist_array(DistGrid%i_strt:,:,:,:)
    integer,           intent(in) :: jref   ! to find sources (defines the row we want)
    logical, optional, intent(in) :: rowcom ! to gather on row_root instead of global root
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: glbl_array(:,:,:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   17 Feb 2015 - Ph. Le Sager - v0
    !
    ! !REMARKS: use in advectx
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'gather_jband_4d_r'

#ifndef MPI
    glbl_array = dist_array( DistGrid%i_strt:DistGrid%i_stop,:,:,:)
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), subarr
    integer :: i0, j0, i1, j1, i0t, j0t, i1t, j1t
    integer :: n, klm, sz(4), gsz(4), slab, tgt_root, stack
    integer(kind=MPI_ADDRESS_KIND) :: sizeoftype, lb, stride
    logical :: selectRoot

    status=0

    ! ------- Check inputs

    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop

    ! by default gather into global root
    selectRoot = isRoot
    tgt_root = root
    if (present(rowcom)) then
       if (rowcom) then
          selectRoot = isRowRoot.and.(jref>=j0).and.(jref<=j1)
          tgt_root = RowRootId
       endif
    endif

    sz = shape(dist_array)
    gsz = shape(glbl_array)
    stack = sz(3)*sz(4)
    
    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, gsz, jband=.true., has_global=selectRoot)
       IF_NOTOK_RETURN(status=1)
    end if

    ! ------- SEND/RECV -------------
    if ( selectRoot ) then

       ! local case
       if((jref>=j0).and.(jref<=j1)) glbl_array(i0:i1,:,:,:) = dist_array(i0:i1,:,:,:)

       ! receive remaining chunks from other pes
       do klm=0,npes-1

          if (klm==myid) cycle  ! skip local case

          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! if klm is a source processor, receive from it
          if((jref<j0t).or.(jref>j1t))cycle
          n=i1t-i0t+1
          call MPI_TYPE_VECTOR (sz(2), n, gsz(1), my_real, slab, ierr)
          CALL MPI_TYPE_GET_EXTENT( my_real, lb, sizeoftype, ierr)
          stride = gsz(1)*gsz(2)*sizeoftype
          call MPI_TYPE_CREATE_HVECTOR (stack, 1, stride, slab, subarr, ierr)
          call MPI_TYPE_FREE (slab, ierr)
          call MPI_TYPE_COMMIT (subarr, ierr)
          call MPI_RECV( glbl_array(i0t,1,1,1), 1, subarr, klm, jref, localComm, stat, ierr)
          call MPI_TYPE_FREE (subarr, ierr)

       end do

    else

       ! are we on a src processor?       
       if((jref<j0).or.(jref>j1))return
       n=i1-i0+1
       call MPI_SEND( dist_array(i0,1,1,1), n*sz(2)*sz(3)*sz(4), my_real, tgt_root, jref, localComm, ierr) 

    end if
#endif
  END SUBROUTINE GATHER_JBAND_4D_R
  !EOC
   
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GATHER_IBAND_1D_R
  !
  ! !DESCRIPTION: gather a meridional (with dimension distributed along J)
  !               vector on root.  For example: [j1:j2]
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GATHER_IBAND_1D_R( DistGrid, dist_array, glbl_array, status, iref )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: dist_array(DistGrid%j_strt:)
    integer,           intent(in) :: iref  ! to define source processors
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: glbl_array(:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS:
    !  - all processors with an i-range containing IREF are used. 
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'gather_iband_1d_r'

#ifndef MPI
    glbl_array = dist_array( DistGrid%j_strt:DistGrid%j_stop )
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), subarr
    integer :: i0, j0, i1, j1, i0t, j0t, i1t, j1t
    integer :: n, klm, sz(1), gsz(1)

    status=0

    ! ------- Check inputs
    sz = shape(dist_array)
    gsz = shape(glbl_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, gsz, iband=.true.)
!       write(gol,*)"lbound m",lbound(dist_array); call goPr
!       write(gol,*)"ubound m",ubound(dist_array); call goPr
       IF_NOTOK_RETURN(status=1)
    end if

    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop

    ! ------- SEND/RECV -------------
    if ( isRoot ) then

       ! local case
       if((iref>=i0).and.(iref<=i1)) glbl_array(j0:j1) = dist_array(j0:j1)

       ! receive remaining chunks from other pes
       do klm=1, npes-1
          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! if klm is a source processor, receive from it
          if((iref<i0t).or.(iref>i1t))cycle
          n=j1t-j0t+1
          call MPI_RECV( glbl_array(j0t), n, my_real, klm, iref, localComm, stat, ierr)
       end do

    else

       ! are we on a src processor?       
       if((iref<i0).or.(iref>i1)) return
       n=j1-j0+1
       call MPI_SEND( dist_array(j0), n, my_real, root, iref, localComm, ierr) 

    end if
#endif
  END SUBROUTINE GATHER_IBAND_1D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    GATHER_IBAND_3D_R
  !
  ! !DESCRIPTION: gather a meridional slice (3D) on root. For arrays like:
  !               [j1:j2, lev, trac], that is without a distributed I dim.
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE GATHER_IBAND_3D_R( DistGrid, dist_array, glbl_array, status, iref )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    real,              intent(in) :: dist_array(DistGrid%j_strt:,:,:)
    integer,           intent(in) :: iref  ! to find sources
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: glbl_array(:,:,:)
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS: use in budget for gathering fluxes
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'gather_iband_3d_r'

#ifndef MPI
    glbl_array = dist_array( DistGrid%j_strt:DistGrid%j_stop,:,:)
    status = 0
#else

    integer :: stat(MPI_STATUS_SIZE), subarr
    integer :: i0, j0, i1, j1, i0t, j0t, i1t, j1t
    integer :: n, klm, sz(3), gsz(3), slab
    integer(kind=MPI_ADDRESS_KIND) :: sizeoftype, lb, stride

    status=0

    ! ------- Check inputs
    sz = shape(dist_array)
    gsz = shape(glbl_array)

    if(okdebug)then
       CALL CHECK_DIST_ARR( DistGrid, sz, 0, status, gsz, iband=.true.)
       IF_NOTOK_RETURN(status=1)
    end if

    i0 = DistGrid%i_strt
    i1 = DistGrid%i_stop
    j0 = DistGrid%j_strt
    j1 = DistGrid%j_stop

    ! ------- SEND/RECV -------------

    if ( isRoot ) then

       ! local case
       if((iref>=i0).and.(iref<=i1)) glbl_array(j0:j1,:,:) = dist_array(j0:j1,:,:)

       ! receive remaining chunks from other pes
       do klm=1,npes-1

          i0t = DistGrid%bounds(1,klm)
          i1t = DistGrid%bounds(2,klm)
          j0t = DistGrid%bounds(3,klm)
          j1t = DistGrid%bounds(4,klm)

          ! if klm is a source processor, receive from it
          if((iref<i0t).or.(iref>i1t))cycle
          n=j1t-j0t+1
          call MPI_TYPE_VECTOR (sz(2), n, gsz(1), my_real, slab, ierr)
          CALL MPI_TYPE_GET_EXTENT( my_real, lb, sizeoftype, ierr)
          stride = gsz(1)*gsz(2)*sizeoftype
          call MPI_TYPE_CREATE_HVECTOR (sz(3), 1, stride, slab, subarr, ierr)
          call MPI_TYPE_FREE (slab, ierr)
          call MPI_TYPE_COMMIT (subarr, ierr)
          call MPI_RECV( glbl_array(j0t,1,1), 1, subarr, klm, iref, localComm, stat, ierr)
          call MPI_TYPE_FREE (subarr, ierr)
       end do
    else
       ! are we on a src processor?
       if((iref<i0).or.(iref>i1))return
       n=j1-j0+1
       call MPI_SEND( dist_array(j0,1,1), n*sz(2)*sz(3), my_real, root, iref, localComm, ierr) 
    end if
#endif
  END SUBROUTINE GATHER_IBAND_3D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    REDUCE_2D_R
  !
  ! !DESCRIPTION: Global reduction. Data are gathered on root, where OP is
  !               then done, instead of OPing on each proc and then calling
  !               MPI_REDUCE. This ensures bitwise agreement with serial code
  !               results in case of SUM. 
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE REDUCE_2D_R( DistGrid, dist_array, halo, op, resultat, status, all, debug)
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    integer,           intent(in) :: halo
    real,              intent(in) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:)
    character(len=3),  intent(in) :: op    ! 'MAX', 'MIN' or 'SUM'
    logical, intent(in), optional :: all   ! mimic mpi_allreduce instead of mpi_reduce
    logical, intent(in), optional :: debug ! print additional information: location of Min/Max
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: resultat
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'REDUCE_2D_R'
    logical :: x_debug
    integer :: shp(2)
    real, allocatable :: glbl_array(:,:)

    x_debug=.false.
    if(present(debug)) x_debug=debug
    
#ifdef MPI

    if (isRoot) then
       allocate( glbl_array( DistGrid%im_region, DistGrid%jm_region ))
    else
       allocate( glbl_array(1,1) )
    end if
    
    call gather(DistGrid, dist_array, glbl_array, halo, status)
    IF_NOTOK_RETURN(status=1)

    if (isRoot) then
       select case( op )

          case('sum', 'Sum', 'SUM')
             resultat = sum(glbl_array)
          case('max', 'Max', 'MAX')
             resultat = maxval(glbl_array)
             if(x_debug) then
                shp=maxloc(glbl_array)
                write(gol,*) rname //": Max location =", shp; call goPr
             end if
          case('min', 'Min', 'MIN')
             resultat = minval(glbl_array)
             if(x_debug) then
                shp=minloc(glbl_array)
                write(gol,*) rname //": Min location =", shp; call goPr
             end if
          case default
             write(gol,*) 'UNSUPPORTED OPERATION'; call goPr; status=1
             IF_NOTOK_RETURN(status=1)

       end select
    end if

    if (present(all)) then
       if (all) call MPI_bcast(resultat, 1, my_real, root, localComm, ierr)
    end if

    deallocate(glbl_array)
#else

    select case( op )
       
       case('sum', 'Sum', 'SUM')
          resultat = sum(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region))
       case('max', 'Max', 'MAX')
          resultat = maxval(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region))
          if(x_debug) then
             shp=maxloc(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region))
             write(gol,*) rname //": Max location =", shp; call goPr
          end if
       case('min', 'Min', 'MIN')
          resultat = minval(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region))
          if(x_debug) then
             shp=minloc(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region))
             write(gol,*) rname //": Min location =", shp; call goPr
          end if          
       case default
          write(gol,*) 'UNSUPPORTED OPERATION'; call goPr; status=1
          IF_NOTOK_RETURN(status=1)

    end select
    
#endif
    
    status=0
       
  END SUBROUTINE REDUCE_2D_R
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    REDUCE_3D_R
  !
  ! !DESCRIPTION: same as REDUCE_2D_R, for 3D arrays. 
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE REDUCE_3D_R( DistGrid, dist_array, halo, op, resultat, status, all, debug)
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    integer,           intent(in) :: halo
    real,              intent(in) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:,:)
    character(len=3),  intent(in) :: op    ! 'MAX', 'MIN' or 'SUM'
    logical, intent(in), optional :: all   ! mimic MPI_ALLREDUCE instead of MPI_REDUCE
    logical, intent(in), optional :: debug ! print additional information: location of Min/Max
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: resultat
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'REDUCE_3D_R'
    integer :: shp(3)
    logical :: x_debug
    real, allocatable :: glbl_array(:,:,:)

    x_debug=.false.
    if(present(debug)) x_debug=debug

#ifdef MPI

    shp = shape( dist_array )

    if (isRoot) then
       allocate( glbl_array( DistGrid%im_region, DistGrid%jm_region, shp(3)) )
    else
       allocate( glbl_array(1,1,1) )
    end if

    call gather(DistGrid, dist_array, glbl_array, halo, status)
    IF_NOTOK_RETURN(status=1)

    if (isRoot) then
       select case( op )

       case('sum', 'Sum', 'SUM')
          resultat = sum(glbl_array)          
       case('max', 'Max', 'MAX')
          resultat = maxval(glbl_array)
          if(x_debug) then
             shp=maxloc(glbl_array)
             write(gol,*) rname //": Max location =", shp; call goPr
          end if          
       case('min', 'Min', 'MIN')
          resultat = minval(glbl_array)          
          if(x_debug) then
             shp=minloc(glbl_array)
             write(gol,*) rname //": Min location =", shp; call goPr
          end if
       case default
          write(gol,*) 'UNSUPPORTED OPERATION'; call goPr; status=1
          IF_NOTOK_RETURN(status=1)
       end select
    end if

    if (present(all)) then
       if (all) call MPI_bcast(resultat, 1, my_real, root, localComm, ierr)
    end if

#else

    select case( op )

    case('sum', 'Sum', 'SUM')
       resultat = sum(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region,:))
    case('max', 'Max', 'MAX')
       resultat = maxval(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region,:))
       if(x_debug) then
          shp=maxloc(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region,:))
          write(gol,*) rname //": Max location =", shp; call goPr
       end if
    case('min', 'Min', 'MIN')
       resultat = minval(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region,:))
       if(x_debug) then
          shp=minloc(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region,:))
          write(gol,*) rname //": Min location =", shp; call goPr
       end if
    case default
       write(gol,*) 'UNSUPPORTED OPERATION'; call goPr; status=1
       IF_NOTOK_RETURN(status=1)

    end select

#endif

    status=0

  END SUBROUTINE REDUCE_3D_R
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    REDUCE_4D_R
  !
  ! !DESCRIPTION: same as REDUCE_2D_R, for 4D arrays.
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE REDUCE_4D_R( DistGrid, dist_array, halo, op, resultat, status, all, debug)
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),   intent(in) :: DistGrid
    integer,           intent(in) :: halo
    real,              intent(in) :: dist_array(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:,:,:)
    character(len=3),  intent(in) :: op    ! 'MAX', 'MIN' or 'SUM'
    logical, intent(in), optional :: all   ! mimic MPI_ALLREDUCE instead of MPI_REDUCE
    logical, intent(in), optional :: debug ! print additional information: location of Min/Max
    !
    ! !OUTPUT PARAMETERS:
    !
    real,    intent(out) :: resultat
    integer, intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'reduce_4d_r'
    integer :: shp(4)
    logical :: x_debug
    real, allocatable :: glbl_array(:,:,:,:)

    x_debug=.false.
    if(present(debug)) x_debug=debug

#ifdef MPI

    shp = shape( dist_array )

    if (isRoot) then
       allocate( glbl_array( DistGrid%im_region, DistGrid%jm_region, shp(3), shp(4)) )
    else
       allocate( glbl_array(1,1,1,1) )
    end if

    call gather(DistGrid, dist_array, glbl_array, halo, status)
    IF_NOTOK_RETURN(status=1)

    if (isRoot) then
       select case( op )

       case('sum', 'Sum', 'SUM')
          resultat = sum(glbl_array)          
       case('max', 'Max', 'MAX')
          resultat = maxval(glbl_array)

          if(x_debug) then
             shp=maxloc(glbl_array)
             write(gol,*) rname //": Max location =", shp; call goPr
          end if

       case('min', 'Min', 'MIN')
          resultat = minval(glbl_array)          

          if(x_debug) then
             shp=minloc(glbl_array)
             write(gol,*) rname //": Min location =", shp; call goPr
          end if

       case default
          write(gol,*) 'UNSUPPORTED OPERATION'; call goPr; status=1
          IF_NOTOK_RETURN(status=1)

       end select
    end if

    if (present(all)) then
       if (all) call MPI_bcast(resultat, 1, my_real, root, localComm, ierr)
    end if

#else

    select case( op )

    case('sum', 'Sum', 'SUM')
       resultat = sum(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region,:,:))
    case('max', 'Max', 'MAX')
       resultat = maxval(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region,:,:))

       if(x_debug) then
          shp=maxloc(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region,:,:))
          write(gol,*) rname //": Max location =", shp; call goPr
       end if

    case('min', 'Min', 'MIN')
       resultat = minval(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region,:,:))

       if(x_debug) then
          shp=minloc(dist_array(1:DistGrid%im_region, 1:DistGrid%jm_region,:,:))
          write(gol,*) rname //": Min location =", shp; call goPr
       end if

    case default
       write(gol,*) 'UNSUPPORTED OPERATION'; call goPr; status=1
       IF_NOTOK_RETURN(status=1)

    end select

#endif

    status=0

  END SUBROUTINE REDUCE_4D_R
  !EOC
  
  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    DIST_ARR_STAT_2D_R
  !
  ! !DESCRIPTION: calculate and print the MIN, MEAN, MAX of a 2D distributed
  !               real field. This is basically a wrapper around several calls
  !               to REDUCE.
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE DIST_ARR_STAT_2D_R( DistGrid, name, arr, halo, status)
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid),  intent(in) :: DistGrid
    character(len=*), intent(in) :: name     ! verbose helper
    integer,          intent(in) :: halo
    real,             intent(in) :: arr(DistGrid%i_strt-halo:,DistGrid%j_strt-halo:)
    !
    ! !OUTPUT PARAMETERS:
    !
    integer,          intent(out):: status
    !
    ! !REVISION HISTORY: 
    !    7 Mar 2012 - P. Le Sager - adapted for lon-lat MPI domain
    !                               decomposition, from DD_FIELD_STATISTICS in 
    !                               both DIFFUSION.F90 and DRY_DEPOSITION.F90    
    !
    ! !REMARKS:
    !    (1) FIXME : does not compute the mean of only non-zero data anymore
    !EOP
    !------------------------------------------------------------------------
    !BOC

    character(len=*), parameter :: rname=mname//'dist_arr_stat_2d_r'
    integer                     :: ntel_non_zero, nx, ny
    real                        :: maxf, minf, meanf, mean_non_zero

    ! --- begin -------------------------------

    call reduce( DistGrid, arr, halo, 'MAX', maxf, status)
    IF_NOTOK_RETURN(status=1)

    call reduce( DistGrid, arr, halo, 'MIN', minf, status)
    IF_NOTOK_RETURN(status=1)

    call reduce( DistGrid, arr, halo, 'SUM', meanf,status)
    IF_NOTOK_RETURN(status=1)

    ! get mean
    nx = DistGrid%im_region
    ny = DistGrid%jm_region
    meanf = meanf / ( nx*ny )

    if(isRoot) then
       write(gol,'(a10,3(a5,1x,1pe10.3))') name,' max=', maxf,' min=',minf,'mean=',meanf!,'mn0',mean_non_zero
       call goPr
    end if
    
    status=0
    
  END SUBROUTINE DIST_ARR_STAT_2D_R
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    PRINT_DISTGRID
  !
  ! !DESCRIPTION: utility that prints information about a distributed grid
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE PRINT_DISTGRID ( DistGrid )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    
    integer, parameter :: maxrow=5
    integer, parameter :: maxcol=5
    integer            :: sz1(1), i1

    ! header
    write(gol,*) "========== Start Distributed Grid ==================="; call goPr

    ! dist_grid
    write(gol,*) "im_region     :" , DistGrid%im_region       ; call goPr
    write(gol,*) "jm_region     :" , DistGrid%jm_region       ; call goPr
    write(gol,*) "i_strt        :" , DistGrid%i_strt          ; call goPr
    write(gol,*) "i_stop        :" , DistGrid%i_stop          ; call goPr
    write(gol,*) "i_strt_halo   :" , DistGrid%i_strt_halo     ; call goPr
    write(gol,*) "i_stop_halo   :" , DistGrid%i_stop_halo     ; call goPr
    write(gol,*) "j_strt        :" , DistGrid%j_strt          ; call goPr
    write(gol,*) "j_stop        :" , DistGrid%j_stop          ; call goPr
    write(gol,*) "j_strt_halo   :" , DistGrid%j_strt_halo     ; call goPr
    write(gol,*) "j_stop_halo   :" , DistGrid%j_stop_halo     ; call goPr
    write(gol,*) "has_north_pole:" , DistGrid%has_north_pole  ; call goPr
    write(gol,*) "has_south_pole:" , DistGrid%has_south_pole  ; call goPr

    ! physical grid
    write(gol,*) '------------- physical grid -------------------' ; call goPr
    write(gol,*) "llgrid name:", trim(DistGrid%lli%name)  ; call goPr
    write(gol,*) "R[m]       :", DistGrid%lli%R           ; call goPr
    write(gol,*) "dlon[deg]  :", DistGrid%lli%dlon_deg    ; call goPr
    write(gol,*) "dlat[deg]  :", DistGrid%lli%dlat_deg    ; call goPr
    write(gol,*) "im         :", DistGrid%lli%im          ; call goPr
    write(gol,*) "jm         :", DistGrid%lli%jm          ; call goPr

    if (associated(DistGrid%lli%lon_deg)) then

       sz1 = shape(DistGrid%lli%lon_deg)
       i1  = min(maxcol,sz1(1))
       write(gol,*) "lon = ",DistGrid%lli%lon_deg(1:i1); call goPr

    endif

    if (associated(DistGrid%lli%lat_deg)) then

       sz1=shape(DistGrid%lli%lat_deg)
       i1=min(maxrow,sz1(1))
       write(gol,*) "lat = ",DistGrid%lli%lat_deg(1:i1); call goPr

    endif

    ! footer
    write(gol,*) "========== End Distributed Grid ===================" ; call goPr


  END SUBROUTINE PRINT_DISTGRID
  !EOC

  !--------------------------------------------------------------------------
  !                    TM5                                                  !
  !--------------------------------------------------------------------------
  !BOP
  !
  ! !IROUTINE:    TESTCOMM
  !
  ! !DESCRIPTION: check if the communications are working as expected.
  !               Currently checked:
  !               - GATHER, SCATTER, UPDATE_HALO (2D, 3D, 4D)
  !               - SCATTER_I_BAND, SCATTER_J_BAND (1D, 2D)
  !\\
  !\\
  ! !INTERFACE:
  !
  SUBROUTINE TESTCOMM( DistGrid, halo, status )
    !
    ! !INPUT PARAMETERS:
    !
    type(dist_grid), intent(in) :: DistGrid
    integer,         intent(in) :: halo
    !
    ! !OUTPUT PARAMETERS:
    !
    integer,        intent(out) :: status
    !
    ! !REVISION HISTORY: 
    !   01 Nov 2011 - P. Le Sager - v0
    !
    ! !REMARKS:
    ! (1) to run with different halo sizes
    ! (2) note that will not catch some errors in halo_update if using too few processors
    !
    !EOP
    !------------------------------------------------------------------------
    !BOC
    character(len=*), parameter  ::  rname = mname//'testcomm'

!     real, allocatable :: src1(:), tgt1(:), res1(:)
!     real, allocatable :: src2(:,:), tgt2(:,:), res2(:,:)
    
!     real, allocatable :: larray2a(:,:),   larray2b(:,:),   glb2a(:,:),   glb2b(:,:)
!     real, allocatable :: larray3a(:,:,:), larray3b(:,:,:), glb3a(:,:,:), glb3b(:,:,:)

!     real, allocatable :: larray4a(:,:,:,:), larray4b(:,:,:,:)
!     real, allocatable ::    glb4a(:,:,:,:),    glb4b(:,:,:,:)

!     integer :: i0, j0, i1, j1, x_halo, k, levels, l, trac, iref(2), jref(2)

!     logical :: south, north, west, east, test
    
!     character(len=*), parameter :: form='(f4.0)'
!     levels=17
!     trac=5

!     ! General
!     call Get_DistGrid( DistGrid, I_STRT=i0, I_STOP=i1, J_STRT=j0, J_STOP=j1, &
!                        hasEastBorder=east,   hasWestBorder=west,             &
!                        hasSouthBorder=south, hasNorthBorder=north            )

!     x_halo=halo
!     status=0

!     if(isRoot) print*, "========= TESTING COMM FOR HALO=",x_HALO

!     call par_barrier()

    
!     ! *************************** SCATTER BAND ***************************

!     iref=(/ 1,  DistGrid%im_region/) ! to test i_band along west and east border
!     jref=(/ 1,  DistGrid%jm_region/) ! to test j_band along south and north border

    
!     if(isRoot) then
!        allocate(src1(DistGrid%im_region))
!     else
!        allocate(src1(1))
!     end if
!     allocate(tgt1(i0:i1), res1(i0:i1))

!     if (isRoot) src1 = (/(k, k=1,DistGrid%im_region)/)
!     res1 = (/(k, k=i0,i1)/)

!     do trac=1,2

!        tgt1=0

!        call scatter_j_band(distgrid, tgt1, src1, status, jref=jref(trac))
!        IF_NOTOK_RETURN(status=1)
       
!        test=((trac==1).and.south).or.((trac==2).and.north)
       
!        ! diff should be 0 along borders only
!        if (maxval(abs(res1-tgt1)) /= 0.) then
!           if(test) then
!              print*, "test scatter_J_band 1D FAILED at border:"
!              !print*, i0,i1,tgt1(i0), tgt1(i1), res1(i0), res1(i1)
!              status=1
!           !else ! Expected only if tgt1 has inout attribute in scatter routine
!           !   print*, "test scatter_J_band 1D PASSED inside:"
!           !   print*, i0,i1,tgt1(i0), tgt1(i1), res1(i0), res1(i1)
!           end if
!        else
!           if(test) then
!              print*, "test scatter_J_band 1D PASSED at border"
!              !print*, i0,i1,tgt1(i0), tgt1(i1), res1(i0), res1(i1)
!           !else ! Expected only if tgt1 has inout attribute in scatter routine
!           !   print*, "test scatter_J_band 1D FAILED inside"
!           !   print*, i0,i1,tgt1(i0), tgt1(i1), res1(i0), res1(i1)
!           !   status=1
!           end if
!        end if
!        IF_NOTOK_RETURN(status=1)
!     end do

!     deallocate(src1, tgt1, res1)


!     if(isRoot) then
!        allocate(src1(DistGrid%jm_region))
!     else
!        allocate(src1(1))
!     end if
!     allocate(tgt1(j0:j1), res1(j0:j1))

!     if (isRoot) src1 = (/(k, k=1,DistGrid%jm_region)/)
!     res1 = (/(k, k=j0,j1)/)

!     do trac=1,2

!        tgt1=0
       
!        call scatter_i_band(distgrid, tgt1, src1, status, iref=iref(trac))
!        IF_NOTOK_RETURN(status=1)

!        test=((trac==1).and.west).or.((trac==2).and.east)

!        ! diff should be 0 along borders only
!        if (maxval(abs(res1-tgt1)) /= 0.) then
!           if(test) then
!              print*, "test scatter_I_band 1D FAILED at border"
!              status=1
!           !else
!           !   print*, "test scatter_I_band 1D PASSED inside"
!           end if
!        else
!           if(test) then
!              print*, "test scatter_I_band 1D PASSED at border"
!           !else
!           !   print*, "test scatter_I_band 1D FAILED inside"
!           !   status=1
!           end if
!        end if
!        IF_NOTOK_RETURN(status=1)

!     end do

!     deallocate(src1, tgt1, res1)

!     ! ---------------- 2D

!     if(isRoot) then
!        allocate(src2(DistGrid%im_region, levels))
!     else
!        allocate(src2(1,1))
!     end if
!     allocate(tgt2(i0:i1,levels), res2(i0:i1,levels))

!     do l=1,levels
!        if (isRoot) src2(:,l) = (/(k, k=1,DistGrid%im_region)/) * l
!        res2(:,l) = (/(k, k=i0,i1)/)* l
!     end do

!     do trac=1,2

!        tgt2=0

!        call scatter_j_band(distgrid, tgt2, src2, status, jref=jref(trac))
!        IF_NOTOK_RETURN(status=1)

!        test=((trac==1).and.south).or.((trac==2).and.north)

!        ! diff should be 0 along borders only
!        if (maxval(abs(res2-tgt2)) /= 0.) then
!           if(test) then
!              print*, "test scatter_J_band 2D FAILED at border"
!              status=1
!           !else
!           !   print*, "test scatter_J_band 2D PASSED inside"
!           end if
!        else
!           if(test) then
!              print*, "test scatter_J_band 2D PASSED at border"
!           !else
!           !   print*, "test scatter_J_band 2D FAILED inside"
!           !   status=1
!           end if
!        end if
!        IF_NOTOK_RETURN(status=1)

!     end do

!     deallocate(src2, tgt2, res2)


!     if(isRoot) then
!        allocate(src2(DistGrid%jm_region,levels))
!     else
!        allocate(src2(1,1))
!     end if
!     allocate(tgt2(j0:j1,levels), res2(j0:j1,levels))

!     do l=1,levels
!        if (isRoot) src2(:,l) = (/(k, k=1,DistGrid%jm_region)/)*l
!        res2(:,l) = (/(k, k=j0,j1)/)*l
!     enddo

!     do trac=1,2

!        tgt2=0

!        call scatter_i_band(distgrid, tgt2, src2, status, iref=iref(trac))
!        IF_NOTOK_RETURN(status=1)

!        test=((trac==1).and.west).or.((trac==2).and.east)

!        ! diff should be 0 along borders only
!        if (maxval(abs(res2-tgt2)) /= 0.) then
!           if(test) then
!              print*, "test scatter_I_band 2D FAILED at border"
!              status=1
!           !else
!           !   print*, "test scatter_I_band 2D PASSED inside"
!           end if
!        else
!           if(test) then
!              print*, "test scatter_I_band 2D PASSED at border"
!           !else
!           !   print*, "test scatter_I_band 2D FAILED inside"
!           !   status=1
!           end if
!        end if
!        IF_NOTOK_RETURN(status=1)

!     end do

!     deallocate(src2, tgt2, res2)


    
!     ! *************************** GATHER/SCATTER ***************************
    
!     !----------------------
!     ! Allocate 2D
!     !----------------------
!     allocate( larray2a( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo) )
!     allocate( larray2b( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo) )

!     allocate( glb2a(DistGrid%im_region, DistGrid%jm_region) )

!     ! in halo, 0, elsewhere myid
!     larray2a=0.
!     larray2a(i0:i1,j0:j1)=real(myid)

!     ! glb2b is global array, used in root only
!     if (isRoot) then 
!        allocate( glb2b( DistGrid%im_region, DistGrid%jm_region) )
!     else
!        allocate( glb2b(1,1) )
!     end if

!     !----------------------
!     ! test GATHER_2D_R
!     !----------------------    
!     glb2b=0.
!     larray2b=0.

!     ! gather
!     call gather( DistGrid, larray2a, glb2b,  x_halo, status)
!     IF_NOTOK_RETURN(status=1)

!     ! broadcast result
!     if (isRoot) glb2a = glb2b    
! #ifdef MPI
!     call MPI_bcast(glb2a, size(glb2a), my_real, root, localComm, ierr)
! #endif

!     larray2b(i0:i1,j0:j1) = glb2a(i0:i1,j0:j1)
!     larray2b = larray2a-larray2b

!     ! diff should be 0
!     if (maxval(abs(larray2b)) /= 0.) then
!        print*, "test gather 2d FAILED"
!        status=1
!     else
!        print*, "test gather 2d PASSED"
!     end if
!     IF_NOTOK_RETURN(status=1)
    
!     call par_barrier()

!     !----------------------
!     ! test SCATTER_2D_R
!     !----------------------
!     larray2b=0.
!     call scatter( DistGrid, larray2b, glb2b,  x_halo, status)
!     IF_NOTOK_RETURN(status=1)

!     larray2b=larray2a-larray2b

!     ! diff should be 0
!     if (maxval(abs(larray2b)) /= 0.) then
!        print*, "test scatter 2d FAILED"
!        status=1
!     else
!        print*, "test scatter 2d PASSED"
!     end if
!     IF_NOTOK_RETURN(status=1)

!     ! CLEANUP
!     deallocate(larray2a,larray2b,glb2a,glb2b)

!     call par_barrier()
    
!     !----------------------
!     ! Allocate 3D
!     !----------------------
!     allocate( larray3a( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo, levels) )
!     allocate( larray3b( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo, levels) )
!     allocate(    glb3a( DistGrid%im_region, DistGrid%jm_region, levels) )

!     ! in halo, 0, elsewhere myid*level
!     larray3a=0.
!     do k=1,levels
!        larray3a(i0:i1,j0:j1,k)=real(myid*k)
!     end do

!     ! glb2b is global array, used in root only
!     if (isRoot) then 
!        allocate( glb3b( DistGrid%im_region, DistGrid%jm_region, levels) )
!     else
!        allocate( glb3b(1,1,1) )
!     end if

!     !----------------------
!     ! test GATHER_3D_R
!     !----------------------    
!     glb3b=0.
!     larray3b=0.

!     ! gather
!     call gather( DistGrid, larray3a, glb3b,  x_halo, status)
!     IF_NOTOK_RETURN(status=1)

!     ! broadcast result
!     if (isRoot) glb3a = glb3b    
! #ifdef MPI
!     call MPI_bcast(glb3a, size(glb3a), my_real, root, localComm, ierr)
! #endif

!     larray3b(i0:i1,j0:j1,:) = glb3a(i0:i1,j0:j1,:)
!     larray3b = larray3a-larray3b

!     ! diff should be 0
!     if (maxval(abs(larray3b)) /= 0.) then
!        print*, "test gather 3d FAILED"
!        status=1
!     else
!        print*, "test gather 3d PASSED"
!     end if
!     IF_NOTOK_RETURN(status=1)

!     call par_barrier()
    
!     !----------------------
!     ! test SCATTER_3D_R
!     !----------------------
!     larray3b=0.
!     call scatter( DistGrid, larray3b, glb3b,  x_halo, status)
!     IF_NOTOK_RETURN(status=1)

!     larray3b=larray3a-larray3b

!     ! diff should be 0
!     if (maxval(abs(larray3b)) /= 0.) then
!        print*, "test scatter 3d FAILED"
!        status=1
!     else
!        print*, "test scatter 3d PASSED"
!     end if
!     IF_NOTOK_RETURN(status=1)

!     ! CLEANUP
!     deallocate(larray3a,larray3b,glb3a,glb3b)

!     call par_barrier()

!     !----------------------
!     ! Allocate 4D
!     !----------------------
!     allocate( larray4a( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo, levels, trac) )
!     allocate( larray4b( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo, levels, trac) )
!     allocate(    glb4a( DistGrid%im_region, DistGrid%jm_region, levels, trac) )

!     ! in halo, 0, elsewhere (myid+1)*level*trac
!     larray4a=0.
!     do l=1,trac
!     do k=1,levels
!        larray4a(i0:i1,j0:j1,k,l)=real((myid+1)*k*l)
!     end do
!     end do
 
!     ! glb4b is global array, used in root only
!     if (isRoot) then 
!        allocate( glb4b( DistGrid%im_region, DistGrid%jm_region, levels, trac) )
!     else
!        allocate( glb4b(1,1,1,1) )
!     end if

!     !----------------------
!     ! test GATHER_4D_R
!     !----------------------    
!     glb4b=0.
!     larray4b=0.

!     ! gather
!     call gather( DistGrid, larray4a, glb4b,  x_halo, status)
!     IF_NOTOK_RETURN(status=1)

!     ! broadcast result
!     if (isRoot) glb4a = glb4b    
! #ifdef MPI
!     call MPI_bcast(glb4a, size(glb4a), my_real, root, localComm, ierr)
! #endif

!     larray4b(i0:i1,j0:j1,:,:) = glb4a(i0:i1,j0:j1,:,:)
!     larray4b = larray4a-larray4b

!     ! diff should be 0
!     if (maxval(abs(larray4b)) /= 0.) then
!        print*, "test gather 4d FAILED"
!        status=1
!     else
!        print*, "test gather 4d PASSED"
!     end if
!     IF_NOTOK_RETURN(status=1)
    
!     call par_barrier()
    
!     !----------------------
!     ! test SCATTER_4D_R
!     !----------------------
!     larray4b=0.
!     call scatter( DistGrid, larray4b, glb4b,  x_halo, status)
!     IF_NOTOK_RETURN(status=1)

!     larray4b=larray4a-larray4b

!     ! diff should be 0
!     if (maxval(abs(larray4b)) /= 0.) then
!        print*, "test scatter 4d FAILED"
!        status=1
!     else
!        print*, "test scatter 4d PASSED"
!     end if
!     IF_NOTOK_RETURN(status=1)

!     ! CLEANUP
!     deallocate(larray4a,larray4b,glb4a,glb4b)
    
!     call par_barrier()
    
!     ! *************************************** HALO ***************************
    
!     !----------------------
!     ! test UPDATE_HALO_2D_R
!     !----------------------
!     ! Allocate 2D
!     allocate( larray2a( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo) )
!     allocate( larray2b( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo) )

!     ! array to update : in halo set to 0, elsewhere to myid
!     larray2b=0.
!     larray2b(i0:i1,j0:j1)=real(myid)

!     ! test array : fill halo with neighbors rank
!     larray2a=0.
!     larray2a( i0-x_halo:i0-1, j0:j1 ) = DistGrid%neighbors(1) ! west halo
!     larray2a( i1+1:i1+x_halo, j0:j1 ) = DistGrid%neighbors(3) ! east halo
!     larray2a( i0:i1, j1+1:j1+x_halo ) = DistGrid%neighbors(2) ! north halo
!     larray2a( i0:i1, j0-x_halo:j0-1 ) = DistGrid%neighbors(4) ! south halo       

!     larray2a(i0:i1,j0:j1)=real(myid)

!     where (larray2a == MPI_PROC_NULL) larray2a=0.
    
!     ! update
!     CALL UPDATE_HALO( DISTGRID,  larray2b,  x_halo, status)
!     IF_NOTOK_RETURN(status=1)

!     if (isRoot.and.(x_halo==1)) then ! 32 is size to look good for 2x2 processes and halo=1
!        print*, "----------------------------"
!        print '(32F4.0)', larray2a
!        print*, "----------------------------"
!        print '(32F4.0)', larray2b
!        print*, "----------------------------"
!     end if

!     ! compare (diff should be 0)    
!     larray2b=larray2a-larray2b

!     if (maxval(abs(larray2b)) /= 0.) then
!        print*, "test update_halo 2d FAILED"
!        status=1
!     else
!        print*, "test update_halo 2d PASSED"
!     end if
!     IF_NOTOK_RETURN(status=1)    

!     ! CLEANUP
!     deallocate(larray2a,larray2b)

!     call par_barrier()
    
!     !----------------------
!     ! test UPDATE_HALO_3D_R
!     !----------------------
!     ! Allocate 3D
!     allocate( larray3a( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo, levels) )
!     allocate( larray3b( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo, levels) )

!     ! array to update : in halo set to 0, elsewhere to myid
!     larray3b=0.
!     larray3b(i0:i1,j0:j1,:)=real(myid)

!     ! test array : fill halo with neighbors rank
!     larray3a=0.
!     larray3a( i0-x_halo:i0-1, j0:j1, : ) = DistGrid%neighbors(1) ! west halo
!     larray3a( i1+1:i1+x_halo, j0:j1, : ) = DistGrid%neighbors(3) ! east halo
!     larray3a( i0:i1, j1+1:j1+x_halo, : ) = DistGrid%neighbors(2) ! north halo
!     larray3a( i0:i1, j0-x_halo:j0-1, : ) = DistGrid%neighbors(4) ! south halo       

!     larray3a(i0:i1,j0:j1,:)=real(myid) ! interior

!     where (larray3a == MPI_PROC_NULL) larray3a=0. !if no neighbor

!     ! update
!     CALL UPDATE_HALO( DISTGRID,  larray3b,  x_halo, status)
!     IF_NOTOK_RETURN(status=1)

!     ! compare (diff should be 0)    
!     larray3b=larray3a-larray3b
    
!     if (maxval(abs(larray3b)) /= 0.) then
!        print*, "test update_halo 3d FAILED"
!        status=1
!     else
!        print*, "test update_halo 3d PASSED"
!     end if
!     IF_NOTOK_RETURN(status=1)
    
!     ! CLEANUP
!     deallocate(larray3a,larray3b)
    
!     call par_barrier()
    
!     !----------------------
!     ! test UPDATE_HALO_4D_R
!     !----------------------
!     ! Allocate 4D
!     allocate( larray4a( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo, levels, trac) )
!     allocate( larray4b( i0-x_halo:i1+x_halo, j0-x_halo:j1+x_halo, levels, trac) )

!     ! array to update : in halo set to 0, elsewhere to myid
!     larray4b=0.
!     larray4b(i0:i1,j0:j1,:,:)=real(myid)

!     ! test array : fill halo with neighbors rank
!     larray4a=0.
!     larray4a( i0-x_halo:i0-1, j0:j1, :,: ) = DistGrid%neighbors(1) ! west halo
!     larray4a( i1+1:i1+x_halo, j0:j1, :,: ) = DistGrid%neighbors(3) ! east halo
!     larray4a( i0:i1, j1+1:j1+x_halo, :,: ) = DistGrid%neighbors(2) ! north halo
!     larray4a( i0:i1, j0-x_halo:j0-1, :,: ) = DistGrid%neighbors(4) ! south halo       

!     larray4a(i0:i1,j0:j1,:,:)=real(myid) ! interior

!     where (larray4a == MPI_PROC_NULL) larray4a=0. !if no neighbor

!     ! update
!     CALL UPDATE_HALO( DISTGRID,  larray4b,  x_halo, status)
!     IF_NOTOK_RETURN(status=1)

!     ! compare (diff should be 0)    
!     larray4b=larray4a-larray4b

!     if (maxval(abs(larray4b)) /= 0.) then
!        print*, "test update_halo 4d FAILED"
!        status=1
!     else
!        print*, "test update_halo 4d PASSED"
!     end if
!     IF_NOTOK_RETURN(status=1)

!     ! CLEANUP
!     deallocate(larray4a,larray4b)
    
!     call par_barrier()

    status=0
    
  END SUBROUTINE TESTCOMM
  !EOC

END MODULE DOMAIN_DECOMP