PRACE NPT

#include <stdio.h>

#include <stdlib.h>

#include "mpi.h"

#include "fundefs.h"

#define  MAXNOD 1000000

int me, nodes;

int offset[MAXNOD][2], sizes[MAXNOD][2];

int main( int argc, char* argv[] )

{ 

   MPI_Comm comm = MPI_COMM_WORLD;

   int      m, l, n;

   int      gok, ok, nrep;

   int      i;

   int      hsize, vsize;

   double   **a, **b, **c;

   double   corr, gtime, nops, speed, time;

   FILE     *inl;

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

   MPI_Init( &argc, &argv );

   MPI_Comm_rank( comm, &me );

   MPI_Comm_size( comm, &nodes );

   if ( me == 0 ) {

      state( "mod2a" );

      prthead( nodes );

   }

   inl = fopen( "mod2am.in", "r" );

   while( ( fscanf( inl, "%d%d%d%d\n", &m, &l, &n, &nrep ) != EOF ) ) {

      sizoff( m, n );

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

// --- Check whether No. of processes <= than No. of matrix rows/cols.

      if ( m < nodes) {

         if ( me == 0 ) {

            printf( "Stop: No. of processes <= than No. of matrix rows.\n" );

            printf( "Increase problem size\n" );

         }

         MPI_Finalize();

         exit( -1 );

      }

      if ( n < nodes) {

         if ( me == 0 ) {

            printf( "Stop: No. of processes <= than No. of matrix columns.\n" );

            printf( "Increase problem size\n" );

         }

         MPI_Finalize();

         exit( -1 );

      }

      vsize = sizes[nodes-1][0];

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

// --- Partition Matrix and allocate.

      hsize = sizes[nodes-1][1];

      a = makmat( vsize, l );

      b = makmat( l, hsize );

      c = makmat( m, hsize );

      gendat( vsize, l, hsize, a, b );

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

// --- Do timing on 'nrep' times repeated matrix-matrix multiply..

      time = MPI_Wtime();

      for( i = 0; i < nrep; i++ ) {

         clr_arr( m, hsize, c );

         gmxm( l, vsize, l, hsize, a, b, c ); 

      }

      time = MPI_Wtime() - time;

      ok   = check( m, l, sizes[me][1], c );

      corr = MPI_Wtime();

      for( i = 0; i < nrep; i++ ) {

         clr_arr( m, hsize, c );

      }

      corr = MPI_Wtime() - corr;

      time = ( time - corr )/nrep;

      nops  = 2*(double)m*(double)l*(double)n;

      MPI_Reduce( &ok, &gok, 1, MPI_INT, MPI_LAND, 0, comm );

      MPI_Reduce( &time, &gtime, 1, MPI_DOUBLE, MPI_MAX, 0, comm );

      if( me == 0 ) prtspeed( m, l, n, gtime, gok, nops );

      delmat( vsize, a );

      delmat( l, b );

      delmat( m, c );

   }

   if ( me == 0 ) {

      printf( "-------------------------------------------------\n" );

   }

   MPI_Finalize();

}