simple run of SPRNG library

sample MPI/c++ program to call the SPRNG library

The full program is runsprng.cpp. Lines illustrating use of the sprng library and MPI Send and Recv functions are shown below.

Using the SPRNG library

#include "runsprng.h"

    /* runsprng.h */
    #define SIMPLE_SPRNG //simple interface to sprng
    #define USE_MPI      //use MPI in sprng
    #include "/home/psiders/include/sprng.h"  //SPRNG header file must be included
    #define SEED 985456376 //seed recommended in sprng 1.0

Initialize a random number stream using generator "gen"

      int gen = 0 ; //0 through 4 selects generator
      init_sprng(gen,SEED,SPRNG_DEFAULT);    // initialize stream

Draw a random number from the stream

      double rannum ;
      ....
      rannum = sprng() ;

Using the MPI functions Send and Recv

define message tags

  #define REPORTRanVal  3
  #define REPORTRanAvg  5
  
  int main () {

initialize MPI and get rank and size

Worker nodes take random numbers from sprng and then send them to the server (node 0) to be collected and printed.
The MPI::COMM_WORLD.Send command sends results to the server.

      MPI::COMM_WORLD.Send(RanValProc, numrans, MPI::DOUBLE, server, REPORTRanVal) ;
      MPI::COMM_WORLD.Send(&RanAvgProc,  1,     MPI::DOUBLE, server, REPORTRanAvg ) ;
RanValProc is a double-precision vector of length numrans.
RanAvgProc is a double-precision scalar.
"server" (=0, int) specifies the destination of the Send.
REPORTRanVal and REPORTRanAVg are integer "tag"s that server will match in the Recv function.

The server waits to receive data from all workers

      int nrecvd = 0 ;  //number of workers from which data have been received
      while (nrecvd < numworkers) {
        MPI::COMM_WORLD.Recv(RanValProc,  numrans, MPI::DOUBLE, MPI_ANY_SOURCE, 
           REPORTRanVal, status) ;
        sender = status.Get_source() ;
         ...
        nrecvd++ ;
      }
      /*These receives loop through the workers in numerical order*/
      for (int worker=1; worker<numprocs; worker++) {
        MPI::COMM_WORLD.Recv(&RanAvgProc, 1, MPI::DOUBLE, worker, 
            REPORTRanAvg, status) ;
        RanAverage[worker] = RanAvgProc ;
      }

The first loop receives data in the order it is sent.
"MPI_ANY_SOURCE" is defined in MPI and means to take data from any sending process.
"status.Get_source();" returns (as int) the rank of the sender
The second loop specifies the sender (=worker) in the Recv function and does not use the status.
"RanAvgProc" is a double-precision scalar.

Output from the command [psiders@beowulf www]$ mpirun -np 6 runsprng

   Random number generator 0 (lfg)
   id:             1          2          3          4          5    
           0    0.44148    0.81419    0.67244    0.06508    0.04786
       20000    0.99797    0.78137    0.27664    0.50201    0.41230
       40000    0.39637    0.38296    0.86886    0.23288    0.71488
       60000    0.40581    0.47516    0.25676    0.11400    0.59770
       80000    0.00248    0.64053    0.95231    0.38301    0.10299
      100000    0.44026    0.31129    0.40963    0.80981    0.41615
   average      0.49995    0.49842    0.49934    0.49823    0.49905
   End of random number test output.
   0.047488 seconds elapsed on 5 workers for generator 0 (lfg).
   ........

Complete output is available in runsprng.log.

go back to SPRNG
go back to index