PRACE NPT

      Subroutine gtrans( a, at, n1, n2, size1, size2, ctime )

! ----------------------------------------------------------------------

! --- 'gtrans' does a global transpose of array 'a' and puts it in 

!     array 'at'.

! ----------------------------------------------------------------------

      Use         numerics

      Use         dist_module ! Contains # of proc.s & proc. no.s

      Implicit    None

      Include     'mpif.h'

      Integer  :: n1, n2, size1, size2

      Real(l_) :: a(n1,size2), at(n2,size1)

      Real(l_) :: ctime

      Real(l_) :: wrk(size2,n1)

      Real(l_) :: time

      Integer  :: actsiz(0:nodes-1,2), base(0:nodes-1,2)

      Integer  :: scnts(0:nodes-1), sdpls(0:nodes-1),

     &            rcnts(0:nodes-1), rdpls(0:nodes-1)

      Integer  :: ierr

! ----------------------------------------------------------------------

! --- Get distribution & offsets for a.

      Call sizoff( n1, n2, actsiz, base )

! ----------------------------------------------------------------------

! --- Do local transposition.

      Call ltrans( n1, actsiz(me,2), a, at )

! ----------------------------------------------------------------------

! --- Determine sizes and displacements of data to be sent.

      Call cntdpls( actsiz, scnts, sdpls, rcnts, rdpls )

! ----------------------------------------------------------------------

! --- Distribute appropriate blocks over the processors.

      time = MPI_Wtime()

      Call MPI_AlltoAllv( at, scnts, sdpls, MPI_Real8,

     &                   wrk, rcnts, rdpls, MPI_Real8,

     &                   MPI_Comm_World, ierr )

      ctime = ctime + MPI_Wtime() - time

! ----------------------------------------------------------------------

! --- Do block transposition of the communicated data.

      Call btrans( wrk, actsiz(me,1), n2, actsiz, at )

! ----------------------------------------------------------------------

      End Subroutine gtrans