SUBROUTINE transpose_xy( f_in, work, f_out )

!--------------------------------------------------------------------------------!
! This file is part of PALM.
!
! PALM is free software: you can redistribute it and/or modify it under the terms
! of the GNU General Public License as published by the Free Software Foundation,
! either version 3 of the License, or (at your option) any later version.
!
! PALM is distributed in the hope that it will be useful, but WITHOUT ANY
! WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
! A PARTICULAR PURPOSE.  See the GNU General Public License for more details.
!
! You should have received a copy of the GNU General Public License along with
! PALM. If not, see <http://www.gnu.org/licenses/>.
!
! Copyright 1997-2012  Leibniz University Hannover
!--------------------------------------------------------------------------------!
!
! Current revisions:
! -----------------
! 
!
! Former revisions:
! -----------------
! $Id: transpose.f90 1093 2013-02-02 12:58:49Z raasch $
!
! 1092 2013-02-02 11:24:22Z raasch
! unused variables removed
!
! 1036 2012-10-22 13:43:42Z raasch
! code put under GPL (PALM 3.9)
!
! 1003 2012-09-14 14:35:53Z raasch
! indices nxa, nya, etc. replaced by nx, ny, etc.
!
! 683 2011-02-09 14:25:15Z raasch
! openMP parallelization of transpositions for 2d-domain-decomposition
!
! 622 2010-12-10 08:08:13Z raasch
! optional barriers included in order to speed up collective operations
!
! 164 2008-05-15 08:46:15Z raasch
! f_inv changed from subroutine argument to automatic array in order to do
! re-ordering from f_in to f_inv in one step, one array work is needed instead
! of work1 and work2
!
! February 2007
! RCS Log replace by Id keyword, revision history cleaned up
!
! Revision 1.2  2004/04/30 13:12:17  raasch
! Switched from mpi_alltoallv to the simpler mpi_alltoall,
! all former transpose-routine files collected in this file, enlarged
! transposition arrays introduced
!
! Revision 1.1  2004/04/30 13:08:16  raasch
! Initial revision (collection of former routines transpose_xy, transpose_xz,
!                   transpose_yx, transpose_yz, transpose_zx, transpose_zy)
!
! Revision 1.1  1997/07/24 11:25:18  raasch
! Initial revision
!
!
! Description:
! ------------
! Transposition of input array (f_in) from x to y. For the input array, all
! elements along x reside on the same PE, while after transposition, all
! elements along y reside on the same PE.
!------------------------------------------------------------------------------!

    USE cpulog
    USE indices
    USE interfaces
    USE pegrid
    USE transpose_indices

    IMPLICIT NONE

    INTEGER ::  i, j, k, l, m, ys
    
    REAL ::  f_in(0:nx,nys_x:nyn_x,nzb_x:nzt_x),   &
             f_inv(nys_x:nyn_x,nzb_x:nzt_x,0:nx),  &
             f_out(0:ny,nxl_y:nxr_y,nzb_y:nzt_y),  &
             work(nnx*nny*nnz)

#if defined( __parallel )

!
!-- Rearrange indices of input array in order to make data to be send
!-- by MPI contiguous
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
    DO  i = 0, nx
       DO  k = nzb_x, nzt_x
          DO  j = nys_x, nyn_x
             f_inv(j,k,i) = f_in(i,j,k)
          ENDDO
       ENDDO
    ENDDO
!$OMP  END PARALLEL

!
!-- Transpose array
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' )
    IF ( collective_wait )  CALL MPI_BARRIER( comm2d, ierr )
    CALL MPI_ALLTOALL( f_inv(nys_x,nzb_x,0), sendrecvcount_xy, MPI_REAL, &
                       work(1),              sendrecvcount_xy, MPI_REAL, &
                       comm1dy, ierr )
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' )

!
!-- Reorder transposed array
!$OMP  PARALLEL PRIVATE ( i, j, k, l, m, ys )
!$OMP  DO
    DO  l = 0, pdims(2) - 1
       m  = l * ( nxr_y - nxl_y + 1 ) * ( nzt_y - nzb_y + 1 ) * &
                ( nyn_x - nys_x + 1 )
       ys = 0 + l * ( nyn_x - nys_x + 1 )
       DO  i = nxl_y, nxr_y
          DO  k = nzb_y, nzt_y
             DO  j = ys, ys + nyn_x - nys_x
                m = m + 1
                f_out(j,i,k) = work(m)
             ENDDO
          ENDDO
       ENDDO
    ENDDO
!$OMP  END PARALLEL

#endif

 END SUBROUTINE transpose_xy


 SUBROUTINE transpose_xz( f_in, work, f_out )

!------------------------------------------------------------------------------!
! Description:
! ------------
! Transposition of input array (f_in) from x to z. For the input array, all
! elements along x reside on the same PE, while after transposition, all
! elements along z reside on the same PE.
!------------------------------------------------------------------------------!

    USE cpulog
    USE indices
    USE interfaces
    USE pegrid
    USE transpose_indices

    IMPLICIT NONE

    INTEGER ::  i, j, k, l, m, xs
    
    REAL ::  f_in(0:nx,nys_x:nyn_x,nzb_x:nzt_x),  &
             f_inv(nys:nyn,nxl:nxr,1:nz),         &
             f_out(1:nz,nys:nyn,nxl:nxr),         &
             work(nnx*nny*nnz)

#if defined( __parallel )

!
!-- If the PE grid is one-dimensional along y, the array has only to be
!-- reordered locally and therefore no transposition has to be done.
    IF ( pdims(1) /= 1 )  THEN
!
!--    Reorder input array for transposition
!$OMP  PARALLEL PRIVATE ( i, j, k, l, m, xs )
!$OMP  DO
       DO  l = 0, pdims(1) - 1
          m  = l * ( nzt_x - nzb_x + 1 ) * nnx * ( nyn_x - nys_x + 1 )
          xs = 0 + l * nnx
          DO  k = nzb_x, nzt_x
             DO  i = xs, xs + nnx - 1
                DO  j = nys_x, nyn_x
                   m = m + 1
                   work(m) = f_in(i,j,k)
                ENDDO
             ENDDO
          ENDDO
       ENDDO
!$OMP  END PARALLEL

!
!--    Transpose array
       CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' )
       IF ( collective_wait )  CALL MPI_BARRIER( comm2d, ierr )
       CALL MPI_ALLTOALL( work(1),          sendrecvcount_zx, MPI_REAL, &
                          f_inv(nys,nxl,1), sendrecvcount_zx, MPI_REAL, &
                          comm1dx, ierr )
       CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' )

!
!--    Reorder transposed array in a way that the z index is in first position
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
       DO  k = 1, nz
          DO  i = nxl, nxr
             DO  j = nys, nyn
                f_out(k,j,i) = f_inv(j,i,k)
             ENDDO
          ENDDO
       ENDDO
!$OMP  END PARALLEL
    ELSE
!
!--    Reorder the array in a way that the z index is in first position
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
       DO  i = nxl, nxr
          DO  j = nys, nyn
             DO  k = 1, nz
                f_inv(j,i,k) = f_in(i,j,k)
             ENDDO
          ENDDO
       ENDDO
!$OMP  END PARALLEL

!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
       DO  k = 1, nz
          DO  i = nxl, nxr
             DO  j = nys, nyn
                f_out(k,j,i) = f_inv(j,i,k)
             ENDDO
          ENDDO
       ENDDO
!$OMP  END PARALLEL

    ENDIF


#endif

 END SUBROUTINE transpose_xz


 SUBROUTINE transpose_yx( f_in, work, f_out )

!------------------------------------------------------------------------------!
! Description:
! ------------
! Transposition of input array (f_in) from y to x. For the input array, all
! elements along y reside on the same PE, while after transposition, all
! elements along x reside on the same PE.
!------------------------------------------------------------------------------!

    USE cpulog
    USE indices
    USE interfaces
    USE pegrid
    USE transpose_indices

    IMPLICIT NONE

    INTEGER ::  i, j, k, l, m, ys
    
    REAL ::  f_in(0:ny,nxl_y:nxr_y,nzb_y:nzt_y),  &
             f_inv(nys_x:nyn_x,nzb_x:nzt_x,0:nx), &
             f_out(0:nx,nys_x:nyn_x,nzb_x:nzt_x), &
             work(nnx*nny*nnz)

#if defined( __parallel )

!
!-- Reorder input array for transposition
!$OMP  PARALLEL PRIVATE ( i, j, k, l, m, ys )
!$OMP  DO
    DO  l = 0, pdims(2) - 1
       m  = l * ( nxr_y - nxl_y + 1 ) * ( nzt_y - nzb_y + 1 ) * &
                ( nyn_x - nys_x + 1 )
       ys = 0 + l * ( nyn_x - nys_x + 1 )
       DO  i = nxl_y, nxr_y
          DO  k = nzb_y, nzt_y
             DO  j = ys, ys + nyn_x - nys_x
                m = m + 1
                work(m) = f_in(j,i,k)
             ENDDO
          ENDDO
       ENDDO
    ENDDO
!$OMP  END PARALLEL

!
!-- Transpose array
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' )
    IF ( collective_wait )  CALL MPI_BARRIER( comm2d, ierr )
    CALL MPI_ALLTOALL( work(1),              sendrecvcount_xy, MPI_REAL, &
                       f_inv(nys_x,nzb_x,0), sendrecvcount_xy, MPI_REAL, &
                       comm1dy, ierr )
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' )

!
!-- Reorder transposed array in a way that the x index is in first position
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
    DO  i = 0, nx
       DO  k = nzb_x, nzt_x
          DO  j = nys_x, nyn_x
             f_out(i,j,k) = f_inv(j,k,i)
          ENDDO
       ENDDO
    ENDDO
!$OMP  END PARALLEL

#endif

 END SUBROUTINE transpose_yx


 SUBROUTINE transpose_yxd( f_in, work, f_out )

!------------------------------------------------------------------------------!
! Description:
! ------------
! Transposition of input array (f_in) from y to x. For the input array, all
! elements along y reside on the same PE, while after transposition, all
! elements along x reside on the same PE.
! This is a direct transposition for arrays with indices in regular order
! (k,j,i) (cf. transpose_yx).
!------------------------------------------------------------------------------!

    USE cpulog
    USE indices
    USE interfaces
    USE pegrid
    USE transpose_indices

    IMPLICIT NONE

    INTEGER ::  i, j, k, l, m, xs

    REAL ::  f_in(1:nz,nys:nyn,nxl:nxr), f_inv(nxl:nxr,1:nz,nys:nyn), &
             f_out(0:nx,nys_x:nyn_x,nzb_x:nzt_x),                     &
             work(nnx*nny*nnz)

#if defined( __parallel )

!
!-- Rearrange indices of input array in order to make data to be send
!-- by MPI contiguous
    DO  k = 1, nz
       DO  j = nys, nyn
          DO  i = nxl, nxr
             f_inv(i,k,j) = f_in(k,j,i)
          ENDDO
       ENDDO
    ENDDO

!
!-- Transpose array
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' )
    IF ( collective_wait )  CALL MPI_BARRIER( comm2d, ierr )
    CALL MPI_ALLTOALL( f_inv(nxl,1,nys), sendrecvcount_xy, MPI_REAL, &
                       work(1),          sendrecvcount_xy, MPI_REAL, &
                       comm1dx, ierr )
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' )

!
!-- Reorder transposed array
    m = 0
    DO  l = 0, pdims(1) - 1
       xs = 0 + l * nnx
       DO  j = nys_x, nyn_x
          DO  k = 1, nz
             DO  i = xs, xs + nnx - 1
                m = m + 1
                f_out(i,j,k) = work(m)
             ENDDO
          ENDDO
       ENDDO
    ENDDO

#endif

 END SUBROUTINE transpose_yxd


 SUBROUTINE transpose_yz( f_in, work, f_out )

!------------------------------------------------------------------------------!
! Description:
! ------------
! Transposition of input array (f_in) from y to z. For the input array, all
! elements along y reside on the same PE, while after transposition, all
! elements along z reside on the same PE.
!------------------------------------------------------------------------------!

    USE cpulog
    USE indices
    USE interfaces
    USE pegrid
    USE transpose_indices

    IMPLICIT NONE

    INTEGER ::  i, j, k, l, m, zs
    
    REAL ::  f_in(0:ny,nxl_y:nxr_y,nzb_y:nzt_y),  &
             f_inv(nxl_y:nxr_y,nzb_y:nzt_y,0:ny), &
             f_out(nxl_z:nxr_z,nys_z:nyn_z,1:nz), &
             work(nnx*nny*nnz)

#if defined( __parallel )

!
!-- Rearrange indices of input array in order to make data to be send
!-- by MPI contiguous
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
    DO  j = 0, ny
       DO  k = nzb_y, nzt_y
          DO  i = nxl_y, nxr_y
             f_inv(i,k,j) = f_in(j,i,k)
          ENDDO
       ENDDO
    ENDDO
!$OMP  END PARALLEL

!
!-- Move data to different array, because memory location of work1 is
!-- needed further below (work1 = work2).
!-- If the PE grid is one-dimensional along y, only local reordering
!-- of the data is necessary and no transposition has to be done.
    IF ( pdims(1) == 1 )  THEN
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
       DO  j = 0, ny
          DO  k = nzb_y, nzt_y
             DO  i = nxl_y, nxr_y
                f_out(i,j,k) = f_inv(i,k,j)
             ENDDO
          ENDDO
       ENDDO
!$OMP  END PARALLEL
       RETURN
    ENDIF

!
!-- Transpose array
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' )
    IF ( collective_wait )  CALL MPI_BARRIER( comm2d, ierr )
    CALL MPI_ALLTOALL( f_inv(nxl_y,nzb_y,0), sendrecvcount_yz, MPI_REAL, &
                       work(1),              sendrecvcount_yz, MPI_REAL, &
                       comm1dx, ierr )
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' )

!
!-- Reorder transposed array
!$OMP  PARALLEL PRIVATE ( i, j, k, l, m, zs )
!$OMP  DO
    DO  l = 0, pdims(1) - 1
       m  = l * ( nyn_z - nys_z + 1 ) * ( nzt_y - nzb_y + 1 ) * &
                ( nxr_z - nxl_z + 1 )
       zs = 1 + l * ( nzt_y - nzb_y + 1 )
       DO  j = nys_z, nyn_z
          DO  k = zs, zs + nzt_y - nzb_y
             DO  i = nxl_z, nxr_z
                m = m + 1
                f_out(i,j,k) = work(m)
             ENDDO
          ENDDO
       ENDDO
    ENDDO
!$OMP  END PARALLEL

#endif

 END SUBROUTINE transpose_yz


 SUBROUTINE transpose_zx( f_in, work, f_out )

!------------------------------------------------------------------------------!
! Description:
! ------------
! Transposition of input array (f_in) from z to x. For the input array, all
! elements along z reside on the same PE, while after transposition, all
! elements along x reside on the same PE.
!------------------------------------------------------------------------------!

    USE cpulog
    USE indices
    USE interfaces
    USE pegrid
    USE transpose_indices

    IMPLICIT NONE

    INTEGER ::  i, j, k, l, m, xs
    
    REAL ::  f_in(1:nz,nys:nyn,nxl:nxr), f_inv(nys:nyn,nxl:nxr,1:nz), &
             f_out(0:nx,nys_x:nyn_x,nzb_x:nzt_x),                     &
             work(nnx*nny*nnz)

#if defined( __parallel )

!
!-- Rearrange indices of input array in order to make data to be send
!-- by MPI contiguous
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
    DO  k = 1,nz
       DO  i = nxl, nxr
          DO  j = nys, nyn
             f_inv(j,i,k) = f_in(k,j,i)
          ENDDO
       ENDDO
    ENDDO
!$OMP  END PARALLEL

!
!-- Move data to different array, because memory location of work1 is
!-- needed further below (work1 = work2).
!-- If the PE grid is one-dimensional along y, only local reordering
!-- of the data is necessary and no transposition has to be done.
    IF ( pdims(1) == 1 )  THEN
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
       DO  k = 1, nz
          DO  i = nxl, nxr
             DO  j = nys, nyn
                f_out(i,j,k) = f_inv(j,i,k)
             ENDDO
          ENDDO
       ENDDO
!$OMP  END PARALLEL
       RETURN
    ENDIF

!
!-- Transpose array
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' )
    IF ( collective_wait )  CALL MPI_BARRIER( comm2d, ierr )
    CALL MPI_ALLTOALL( f_inv(nys,nxl,1), sendrecvcount_zx, MPI_REAL, &
                       work(1),          sendrecvcount_zx, MPI_REAL, &
                       comm1dx, ierr )
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' )

!
!-- Reorder transposed array
!$OMP  PARALLEL PRIVATE ( i, j, k, l, m, xs )
!$OMP  DO
    DO  l = 0, pdims(1) - 1
       m  = l * ( nzt_x - nzb_x + 1 ) * nnx * ( nyn_x - nys_x + 1 )
       xs = 0 + l * nnx
       DO  k = nzb_x, nzt_x
          DO  i = xs, xs + nnx - 1
             DO  j = nys_x, nyn_x
                m = m + 1
                f_out(i,j,k) = work(m)
             ENDDO
          ENDDO
       ENDDO
    ENDDO
!$OMP  END PARALLEL

#endif

 END SUBROUTINE transpose_zx


 SUBROUTINE transpose_zy( f_in, work, f_out )

!------------------------------------------------------------------------------!
! Description:
! ------------
! Transposition of input array (f_in) from z to y. For the input array, all
! elements along z reside on the same PE, while after transposition, all
! elements along y reside on the same PE.
!------------------------------------------------------------------------------!

    USE cpulog
    USE indices
    USE interfaces
    USE pegrid
    USE transpose_indices

    IMPLICIT NONE

    INTEGER ::  i, j, k, l, m, zs
    
    REAL ::  f_in(nxl_z:nxr_z,nys_z:nyn_z,1:nz),  &
             f_inv(nxl_y:nxr_y,nzb_y:nzt_y,0:ny), &
             f_out(0:ny,nxl_y:nxr_y,nzb_y:nzt_y), &
             work(nnx*nny*nnz)

#if defined( __parallel )

!
!-- If the PE grid is one-dimensional along y, the array has only to be
!-- reordered locally and therefore no transposition has to be done.
    IF ( pdims(1) /= 1 )  THEN
!
!--    Reorder input array for transposition
!$OMP  PARALLEL PRIVATE ( i, j, k, l, m, zs )
!$OMP  DO
       DO  l = 0, pdims(1) - 1
          m  = l * ( nyn_z - nys_z + 1 ) * ( nzt_y - nzb_y + 1 ) * &
                   ( nxr_z - nxl_z + 1 )
          zs = 1 + l * ( nzt_y - nzb_y + 1 )
          DO  j = nys_z, nyn_z
             DO  k = zs, zs + nzt_y - nzb_y
                DO  i = nxl_z, nxr_z
                   m = m + 1
                   work(m) = f_in(i,j,k)
                ENDDO
             ENDDO
          ENDDO
       ENDDO
!$OMP  END PARALLEL

!
!--    Transpose array
       CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' )
       IF ( collective_wait )  CALL MPI_BARRIER( comm2d, ierr )
       CALL MPI_ALLTOALL( work(1),              sendrecvcount_yz, MPI_REAL, &
                          f_inv(nxl_y,nzb_y,0), sendrecvcount_yz, MPI_REAL, &
                          comm1dx, ierr )
       CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' )

!
!--    Reorder transposed array in a way that the y index is in first position
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
       DO  j = 0, ny
          DO  k = nzb_y, nzt_y
             DO  i = nxl_y, nxr_y
                f_out(j,i,k) = f_inv(i,k,j)
             ENDDO
          ENDDO
       ENDDO
!$OMP  END PARALLEL
    ELSE
!
!--    Reorder the array in a way that the y index is in first position
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
       DO  k = nzb_y, nzt_y
          DO  j = 0, ny
             DO  i = nxl_y, nxr_y
                f_inv(i,k,j) = f_in(i,j,k)
             ENDDO
          ENDDO
       ENDDO
!$OMP  END PARALLEL
!
!--    Move data to output array
!$OMP  PARALLEL PRIVATE ( i, j, k )
!$OMP  DO
       DO  k = nzb_y, nzt_y
          DO  i = nxl_y, nxr_y
             DO  j = 0, ny
                f_out(j,i,k) = f_inv(i,k,j)
             ENDDO
          ENDDO
       ENDDO
!$OMP  END PARALLEL

    ENDIF

#endif

 END SUBROUTINE transpose_zy


 SUBROUTINE transpose_zyd( f_in, work, f_out )

!------------------------------------------------------------------------------!
! Description:
! ------------
! Transposition of input array (f_in) from z to y. For the input array, all
! elements along z reside on the same PE, while after transposition, all
! elements along y reside on the same PE.
! This is a direct transposition for arrays with indices in regular order
! (k,j,i) (cf. transpose_zy).
!------------------------------------------------------------------------------!

    USE cpulog
    USE indices
    USE interfaces
    USE pegrid
    USE transpose_indices

    IMPLICIT NONE

    INTEGER ::  i, j, k, l, m, ys
    
    REAL ::  f_in(1:nz,nys:nyn,nxl:nxr), f_inv(nys:nyn,nxl:nxr,1:nz), &
             f_out(0:ny,nxl_yd:nxr_yd,nzb_yd:nzt_yd),                 &
             work(nnx*nny*nnz)

#if defined( __parallel )

!
!-- Rearrange indices of input array in order to make data to be send
!-- by MPI contiguous
    DO  i = nxl, nxr
       DO  j = nys, nyn
          DO  k = 1, nz
             f_inv(j,i,k) = f_in(k,j,i)
          ENDDO
       ENDDO
    ENDDO

!
!-- Move data to different array, because memory location of work1 is
!-- needed further below (work1 = work2).
!-- If the PE grid is one-dimensional along x, only local reordering
!-- of the data is necessary and no transposition has to be done.
    IF ( pdims(2) == 1 )  THEN
       DO  k = 1, nz
          DO  i = nxl, nxr
             DO  j = nys, nyn
                f_out(j,i,k) = f_inv(j,i,k)
             ENDDO
          ENDDO
       ENDDO
       RETURN
    ENDIF

!
!-- Transpose array
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' )
    IF ( collective_wait )  CALL MPI_BARRIER( comm2d, ierr )
    CALL MPI_ALLTOALL( f_inv(nys,nxl,1), sendrecvcount_zyd, MPI_REAL, &
                       work(1),          sendrecvcount_zyd, MPI_REAL, &
                       comm1dy, ierr )
    CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' )

!
!-- Reorder transposed array
    m = 0
    DO  l = 0, pdims(2) - 1
       ys = 0 + l * nny
       DO  k = nzb_yd, nzt_yd
          DO  i = nxl_yd, nxr_yd
             DO  j = ys, ys + nny - 1
                m = m + 1
                f_out(j,i,k) = work(m)
             ENDDO
          ENDDO
       ENDDO
    ENDDO

#endif

 END SUBROUTINE transpose_zyd