BetwCent.cpp

Go to the documentation of this file.

/****************************************************************/
/* Parallel Combinatorial BLAS Library (for Graph Computations) */
/* version 1.6 -------------------------------------------------*/
/* date: 6/15/2017 ---------------------------------------------*/
/* authors: Ariful Azad, Aydin Buluc  --------------------------*/
/****************************************************************/
/*
 Copyright (c) 2010-2017, The Regents of the University of California
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal
 in the Software without restriction, including without limitation the rights
 to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 copies of the Software, and to permit persons to whom the Software is
 furnished to do so, subject to the following conditions:
 
 The above copyright notice and this permission notice shall be included in
 all copies or substantial portions of the Software.
 
 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 THE SOFTWARE.
 */
 
// These macros should be defined before stdint.h is included
#ifndef __STDC_CONSTANT_MACROS
#define __STDC_CONSTANT_MACROS
#endif
#ifndef __STDC_LIMIT_MACROS
#define __STDC_LIMIT_MACROS
#endif
#include <stdint.h>
#include <mpi.h>
#include <iostream>
#include <fstream>
#include <string>
#include <sstream>  // Required for stringstreams
#include <ctime>
#include <cmath>
#include "CombBLAS/CombBLAS.h"
 
using namespace combblas;
using namespace std;
 
// Simple helper class for declarations: Just the numerical type is templated 
// The index type and the sequential matrix type stays the same for the whole code
// In this case, they are "int" and "SpDCCols"
template <class NT>
class Dist 
{ 
public: 
    typedef SpDCCols < int, NT > DCCols;
    typedef SpParMat < int, NT, DCCols > MPI_DCCols;
};
 
 
int main(int argc, char* argv[])
{
    int nprocs, myrank;
    MPI_Init(&argc, &argv);
    MPI_Comm_size(MPI_COMM_WORLD,&nprocs);
    MPI_Comm_rank(MPI_COMM_WORLD,&myrank);
    
    typedef PlusTimesSRing<bool, int> PTBOOLINT;    
    typedef PlusTimesSRing<bool, double> PTBOOLDOUBLE;
 
    if(argc < 4)
    {
        if(myrank == 0)
        {   
                cout << "Usage: ./betwcent <BASEADDRESS> <K4APPROX> <BATCHSIZE> <output file - optional>" << endl;
                cout << "Example: ./betwcent Data/ 15 128" << endl;
                cout << "Input file input.mtx should be under <BASEADDRESS> in matrix market format" << endl;
                cout << "<BATCHSIZE> should be a multiple of sqrt(p)" << endl;
                cout << "Because <BATCHSIZE> is for the overall matrix (similarly, <K4APPROX> is global as well) " << endl;
        }
        MPI_Finalize();
        return -1;
        }
 
    {
        int K4Approx = atoi(argv[2]);
        int batchSize = atoi(argv[3]);
 
        string directory(argv[1]);      
        string ifilename = "input.mtx";
        ifilename = directory+"/"+ifilename;
 
        shared_ptr<CommGrid> fullWorld;
        fullWorld.reset( new CommGrid(MPI_COMM_WORLD, 0, 0) );
        
        Dist<bool>::MPI_DCCols A(fullWorld);
            Dist<bool>::MPI_DCCols AT(fullWorld);   // construct object
        AT.ParallelReadMM(ifilename, true,  maximum<double>()); // read it from file, note that we use the transpose of "input" data
        A = AT;
        A.Transpose();
            
        int nPasses = (int) pow(2.0, K4Approx);
        int numBatches = (int) ceil( static_cast<float>(nPasses)/ static_cast<float>(batchSize));
    
        // get the number of batch vertices for submatrix
        int subBatchSize = batchSize / (AT.getcommgrid())->GetGridCols();
        int nBatchSize = subBatchSize * (AT.getcommgrid())->GetGridCols();
        nPasses = numBatches * nBatchSize;  // update the number of starting vertices   
    
        if(batchSize % (AT.getcommgrid())->GetGridCols() > 0 && myrank == 0)
        {
            cout << "*** Batchsize is not evenly divisible by the grid dimension ***" << endl;
            cout << "*** Processing "<< nPasses <<" vertices instead"<< endl;
        }
 
            A.PrintInfo();
            ostringstream tinfo;
        tinfo << "Batch processing will occur " << numBatches << " times, each processing " << nBatchSize << " vertices (overall)" << endl;
            SpParHelper::Print(tinfo.str());
 
            vector<int> candidates;
        // Only consider non-isolated vertices
        int vertices = 0;
        int vrtxid = 0; 
        int nlocpass = numBatches * subBatchSize;
        while(vertices < nlocpass)
        {
            vector<int> single;
            vector<int> empty;
            single.push_back(vrtxid);       // will return ERROR if vrtxid > N (the column dimension) 
            int locnnz = ((AT.seq())(empty,single)).getnnz();
            int totnnz;
            MPI_Allreduce( &locnnz, &totnnz, 1, MPI_INT, MPI_SUM, (AT.getcommgrid())->GetColWorld());
                    
            if(totnnz > 0)
            {
                candidates.push_back(vrtxid);
                ++vertices;
            }
            ++vrtxid;
        }
        
        SpParHelper::Print("Candidates chosen, precomputation finished\n");
        double t1 = MPI_Wtime();
        vector<int> batch(subBatchSize);
        FullyDistVec<int, double> bc(AT.getcommgrid(), A.getnrow(), 0.0);
 
        for(int i=0; i< numBatches; ++i)
        {
            for(int j=0; j< subBatchSize; ++j)
            {
                batch[j] = candidates[i*subBatchSize + j];
            }
            
            Dist<int>::MPI_DCCols fringe = AT.SubsRefCol(batch);
 
            // Create nsp by setting (r,i)=1 for the ith root vertex with label r
            // Inially only the diagonal processors have any nonzeros (because we chose roots so)
            Dist<int>::DCCols * nsploc = new Dist<int>::DCCols();
            tuple<int, int, int> * mytuples = NULL; 
            if(AT.getcommgrid()->GetRankInProcRow() == AT.getcommgrid()->GetRankInProcCol())
            {
                mytuples = new tuple<int, int, int>[subBatchSize];
                for(int k =0; k<subBatchSize; ++k)
                {
                    mytuples[k] = make_tuple(batch[k], k, 1);
                }
                nsploc->Create( subBatchSize, AT.getlocalrows(), subBatchSize, mytuples);       
            }
            else
            {  
                nsploc->Create( 0, AT.getlocalrows(), subBatchSize, mytuples);      
            }
        
            Dist<int>::MPI_DCCols  nsp(nsploc, AT.getcommgrid());           
            vector < Dist<bool>::MPI_DCCols * > bfs;        // internally keeps track of depth
 
            SpParHelper::Print("Exploring via BFS...\n");
            while( fringe.getnnz() > 0 )
            {
                nsp += fringe;
                Dist<bool>::MPI_DCCols * level = new Dist<bool>::MPI_DCCols( fringe ); 
                bfs.push_back(level);
 
                fringe = PSpGEMM<PTBOOLINT>(AT, fringe);
                fringe = EWiseMult(fringe, nsp, true);
            }
 
            // Apply the unary function 1/x to every element in the matrix
            // 1/x works because no explicit zeros are stored in the sparse matrix nsp
            Dist<double>::MPI_DCCols nspInv = nsp;
            nspInv.Apply(bind1st(divides<double>(), 1));
 
            // create a dense matrix with all 1's 
            DenseParMat<int, double> bcu(1.0, AT.getcommgrid(), fringe.getlocalrows(), fringe.getlocalcols() );
 
            SpParHelper::Print("Tallying...\n");
            // BC update for all vertices except the sources
            for(int j = bfs.size()-1; j > 0; --j)
            {
                Dist<double>::MPI_DCCols w = EWiseMult( *bfs[j], nspInv, false);
                w.EWiseScale(bcu);
 
                Dist<double>::MPI_DCCols product = PSpGEMM<PTBOOLDOUBLE>(A,w);
                product = EWiseMult(product, *bfs[j-1], false);
                product = EWiseMult(product, nsp, false);       
 
                bcu += product;
            }
            for(int j=0; j < bfs.size(); ++j)
            {
                delete bfs[j];
            }
        
            SpParHelper::Print("Adding bc contributions...\n");
            bc += FullyDistVec<int, double>(bcu.Reduce(Row, plus<double>(), 0.0));  // pack along rows
        }
        bc.Apply(bind2nd(minus<double>(), nPasses));    // Subtrack nPasses from all the bc scores (because bcu was initialized to all 1's)
        
        double t2=MPI_Wtime();
        double TEPS = (nPasses * static_cast<float>(A.getnnz())) / (t2-t1);
        if( myrank == 0)
        {
            cout<<"Computation finished"<<endl; 
            fprintf(stdout, "%.6lf seconds elapsed for %d starting vertices\n", t2-t1, nPasses);
            fprintf(stdout, "TEPS score is: %.6lf\n", TEPS);
        }
        ofstream output(argv[4]);
        bc.SaveGathered(output, 0);
        output.close();
    }   
 
    // make sure the destructors for all objects are called before MPI::Finalize()
    MPI_Finalize();
    return 0;
}