PRACE NPT

#include <stdlib.h>

#include <mpi.h>

#include "mpiargs.h"

#include "fundefs.h"

void gtrans( int n1, int n2, int m1, int m2,

             double **ar, double **ai, double **atr, double **ati,

             int dir, double *time )

// ---------------------------------------------------------------------

//    'gtrans' does a global transposition of arrays 'ar' & 'ai' and

//     puts them in arrays 'atr' & 'ati'.

// ---------------------------------------------------------------------

{

   MPI_Comm   comm = MPI_COMM_WORLD;

   double     **wrkr, **wrk2r, **wrki, **wrk2i;

   int        *scnts, *sdpls, *rcnts, *rdpls;

   double     *bufin, *bufout;

   double     ltime;

// ---------------------------------------------------------------------

// --- If nodes = 1, only local transposition.

   if ( nodes == 1 ) {

      ltrans( n1, n2,  ar, ai, atr, ati );

      return;

   }

// ---------------------------------------------------------------------

// --- Make work arrays.

   wrkr  = makmat( n2, m1 ); wrki = makmat( n2, m1 );

   bufin  = calloc( n2*m1, sizeof( double ) );

   bufout = calloc( n1*m2, sizeof( double ) );

   scnts  = calloc( nodes, sizeof( int ) );

   sdpls  = calloc( nodes, sizeof( int ) );

   rcnts  = calloc( nodes, sizeof( int ) );

   rdpls  = calloc( nodes, sizeof( int ) );

// ---------------------------------------------------------------------

// --- Do local transpositions first.

   ltrans( m1, n2,  ar, ai, wrkr, wrki );

// ---------------------------------------------------------------------

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

   cntdpls( scnts, sdpls, rcnts, rdpls, dir );

// ---------------------------------------------------------------------

// --- Distribute appropriate blocks over the processors and do a block

//     transposition on the output buffers afterwards to get the

//     elements of 'at[r|i]' in the right order.

   d2to1( n2, m1, wrkr, bufin );

   ltime = MPI_Wtime();

   MPI_Alltoallv( bufin,  scnts, sdpls, MPI_DOUBLE,

                  bufout, rcnts, rdpls, MPI_DOUBLE, comm );

   btrans( n1, n2, bufout, atr, dir );

   d2to1( n2, m1, wrki, bufin );

   MPI_Alltoallv( bufin,  scnts, sdpls, MPI_DOUBLE,

                  bufout, rcnts, rdpls, MPI_DOUBLE, comm );

   btrans( n1, n2, bufout, ati, dir );

   *time = *time + MPI_Wtime() - ltime;

// ---------------------------------------------------------------------

// --- Clean up.

   free( bufin ); free( bufout );

   free( scnts ); free( sdpls ); free( rcnts ); free( rdpls );

   delmat( m2, wrki ); delmat( m2, wrkr );

}