Sample Java Irregular Program

The Java code presented below shows the irregular out-of-core program that uses the Hilbert partitioner as described in Section . The declaration of lip structures in the Java class is convenient since they are implemented as objects:

 import pl.edu.agh.icsr.lip.*;
 import mpi.*;
 import pl.edu.agh.icsr.janet.tools.*;  
  
 public class Data_ooc_irr {

    /* ... */

    public static void main(String[] args) throws MPIException { 

        /* ... */

        Datamap datamap, filetype, memtype; 
        Schedule schedule; 
        IOBufmap bufm; 
        Datamap datamap_irr; 

        MPI.Init(args);
        LIP.Setup(null, 0);
        info = new mpi.Info();

        rank = LIP.Rank();
        size = LIP.Size();

        datamap = new Datamap(new Datamap.BlockDistribution(l, l_l), 1, 0); 
        datamap_irr = new Datamap();
        datamap_irr.create_hilbert_distribution_ooc(datamap, xfile, yfile, 
             null, l_l, ps, sample, new VDouble(vertex),edge,minedge);

Calling lip methods in Java also comes naturally by using virtual arrays VDouble with the method subArray to reference a portion of array data. It replaces the pointers that are commonly used in C library calls to pass arrays to functions.

        
        mpi.File file_irr = new mpi.File();
        file_irr.open(MPI.COMM_WORLD, "data_irr.file", 
                mpi.File.MODE_RDWR | mpi.File.MODE_CREATE, null);
             
        datamap_irr.remap_ooc(datamap,file,file_irr,l_l, ps,
                MPI.DOUBLE, Datamap.CREATE_LOCAL | Datamap.MOVE_DATA);
        
        bufm = new IOBufmap(l_l);
     
        schedule = new Schedule(l_l);

The index layout is encapsulated in the ContIntIndexer class which eliminates the need for using many flags to localize functions, as was the case in C.

        LIP.Localize(datamap_irr,   
               new LIP.ContIntIndexer(perm),
               new LIP.ContIntIndexer(perm_l),
               0, schedule);
        LIP.OOC_Localize(new LIP.ContIntIndexer(perm_l),
                   new LIP.ContIntIndexer(tperm_l),
                   0, new VDouble(x_l), MPI.DOUBLE, bufm, schedule);
        /* ... */

        LIP.Gather(new VDouble(x_l), new VDouble(x_l).subArray(l_l), 
             MPI.DOUBLE, schedule ); 
        
        /* irregular loops 
        ...
        */
        
        LIP.Scatter(new VDouble(x_l).subArray(l_l), new VDouble(x_l), 
              MPI.DOUBLE, schedule, LIP.OP_SUM);
        
        schedule.free();
        schedule = null;

Created by Katarzyna Zając