ground/gcs/src/libs/eigen/bench/sparse_dense_product.cpp

   1
   2 //g++ -O3 -g0 -DNDEBUG  sparse_product.cpp -I.. -I/home/gael/Coding/LinearAlgebra/mtl4/ -DDENSITY=0.005 -DSIZE=10000 && ./a.out
   3 //g++ -O3 -g0 -DNDEBUG  sparse_product.cpp -I.. -I/home/gael/Coding/LinearAlgebra/mtl4/ -DDENSITY=0.05 -DSIZE=2000 && ./a.out
   4 // -DNOGMM -DNOMTL -DCSPARSE
   5 // -I /home/gael/Coding/LinearAlgebra/CSparse/Include/ /home/gael/Coding/LinearAlgebra/CSparse/Lib/libcsparse.a
   6 #ifndef SIZE
   7 #define SIZE 650000
   8 #endif
   9
  10 #ifndef DENSITY
  11 #define DENSITY 0.01
  12 #endif
  13
  14 #ifndef REPEAT
  15 #define REPEAT 1
  16 #endif
  17
  18 #include "BenchSparseUtil.h"
  19
  20 #ifndef MINDENSITY
  21 #define MINDENSITY 0.0004
  22 #endif
  23
  24 #ifndef NBTRIES
  25 #define NBTRIES 10
  26 #endif
  27
  28 #define BENCH(X) \
  29   timer.reset(); \
  30   for (int _j=0; _j<NBTRIES; ++_j) { \
  31     timer.start(); \
  32     for (int _k=0; _k<REPEAT; ++_k) { \
  33         X  \
  34   } timer.stop(); }
  35
  36
  37 #ifdef CSPARSE
  38 cs* cs_sorted_multiply(const cs* a, const cs* b)
  39 {
  40   cs* A = cs_transpose (a, 1) ;
  41   cs* B = cs_transpose (b, 1) ;
  42   cs* D = cs_multiply (B,A) ;   /* D = B'*A' */
  43   cs_spfree (A) ;
  44   cs_spfree (B) ;
  45   cs_dropzeros (D) ;      /* drop zeros from D */
  46   cs* C = cs_transpose (D, 1) ;   /* C = D', so that C is sorted */
  47   cs_spfree (D) ;
  48   return C;
  49 }
  50 #endif
  51
  52 int main(int argc, char *argv[])
  53 {
  54   int rows = SIZE;
  55   int cols = SIZE;
  56   float density = DENSITY;
  57
  58   EigenSparseMatrix sm1(rows,cols);
  59   DenseVector v1(cols), v2(cols);
  60   v1.setRandom();
  61
  62   BenchTimer timer;
  63   for (float density = DENSITY; density>=MINDENSITY; density*=0.5)
  64   {
  65     //fillMatrix(density, rows, cols, sm1);
  66     fillMatrix2(7, rows, cols, sm1);
  67
  68     // dense matrices
  69     #ifdef DENSEMATRIX
  70     {
  71       std::cout << "Eigen Dense\t" << density*100 << "%\n";
  72       DenseMatrix m1(rows,cols);
  73       eiToDense(sm1, m1);
  74
  75       timer.reset();
  76       timer.start();
  77       for (int k=0; k<REPEAT; ++k)
  78         v2 = m1 * v1;
  79       timer.stop();
  80       std::cout << "   a * v:\t" << timer.best() << "  " << double(REPEAT)/timer.best() << " * / sec " << endl;
  81
  82       timer.reset();
  83       timer.start();
  84       for (int k=0; k<REPEAT; ++k)
  85         v2 = m1.transpose() * v1;
  86       timer.stop();
  87       std::cout << "   a' * v:\t" << timer.best() << endl;
  88     }
  89     #endif
  90
  91     // eigen sparse matrices
  92     {
  93       std::cout << "Eigen sparse\t" << sm1.nonZeros()/float(sm1.rows()*sm1.cols())*100 << "%\n";
  94
  95       BENCH(asm("#myc"); v2 = sm1 * v1; asm("#myd");)
  96       std::cout << "   a * v:\t" << timer.best()/REPEAT << "  " << double(REPEAT)/timer.best(REAL_TIMER) << " * / sec " << endl;
  97
  98
  99       BENCH( { asm("#mya"); v2 = sm1.transpose() * v1; asm("#myb"); })
 100
 101       std::cout << "   a' * v:\t" << timer.best()/REPEAT << endl;
 102     }
 103
 104 //     {
 105 //       DynamicSparseMatrix<Scalar> m1(sm1);
 106 //       std::cout << "Eigen dyn-sparse\t" << m1.nonZeros()/float(m1.rows()*m1.cols())*100 << "%\n";
 107 //
 108 //       BENCH(for (int k=0; k<REPEAT; ++k) v2 = m1 * v1;)
 109 //       std::cout << "   a * v:\t" << timer.value() << endl;
 110 //
 111 //       BENCH(for (int k=0; k<REPEAT; ++k) v2 = m1.transpose() * v1;)
 112 //       std::cout << "   a' * v:\t" << timer.value() << endl;
 113 //     }
 114
 115     // GMM++
 116     #ifndef NOGMM
 117     {
 118       std::cout << "GMM++ sparse\t" << density*100 << "%\n";
 119       //GmmDynSparse  gmmT3(rows,cols);
 120       GmmSparse m1(rows,cols);
 121       eiToGmm(sm1, m1);
 122
 123       std::vector<Scalar> gmmV1(cols), gmmV2(cols);
 124       Map<Matrix<Scalar,Dynamic,1> >(&gmmV1[0], cols) = v1;
 125       Map<Matrix<Scalar,Dynamic,1> >(&gmmV2[0], cols) = v2;
 126
 127       BENCH( asm("#myx"); gmm::mult(m1, gmmV1, gmmV2); asm("#myy"); )
 128       std::cout << "   a * v:\t" << timer.value() << endl;
 129
 130       BENCH( gmm::mult(gmm::transposed(m1), gmmV1, gmmV2); )
 131       std::cout << "   a' * v:\t" << timer.value() << endl;
 132     }
 133     #endif
 134
 135     #ifndef NOUBLAS
 136     {
 137       std::cout << "ublas sparse\t" << density*100 << "%\n";
 138       UBlasSparse m1(rows,cols);
 139       eiToUblas(sm1, m1);
 140
 141       boost::numeric::ublas::vector<Scalar> uv1, uv2;
 142       eiToUblasVec(v1,uv1);
 143       eiToUblasVec(v2,uv2);
 144
 145 //       std::vector<Scalar> gmmV1(cols), gmmV2(cols);
 146 //       Map<Matrix<Scalar,Dynamic,1> >(&gmmV1[0], cols) = v1;
 147 //       Map<Matrix<Scalar,Dynamic,1> >(&gmmV2[0], cols) = v2;
 148
 149       BENCH( uv2 = boost::numeric::ublas::prod(m1, uv1); )
 150       std::cout << "   a * v:\t" << timer.value() << endl;
 151
 152 //       BENCH( boost::ublas::prod(gmm::transposed(m1), gmmV1, gmmV2); )
 153 //       std::cout << "   a' * v:\t" << timer.value() << endl;
 154     }
 155     #endif
 156
 157     // MTL4
 158     #ifndef NOMTL
 159     {
 160       std::cout << "MTL4\t" << density*100 << "%\n";
 161       MtlSparse m1(rows,cols);
 162       eiToMtl(sm1, m1);
 163       mtl::dense_vector<Scalar> mtlV1(cols, 1.0);
 164       mtl::dense_vector<Scalar> mtlV2(cols, 1.0);
 165
 166       timer.reset();
 167       timer.start();
 168       for (int k=0; k<REPEAT; ++k)
 169         mtlV2 = m1 * mtlV1;
 170       timer.stop();
 171       std::cout << "   a * v:\t" << timer.value() << endl;
 172
 173       timer.reset();
 174       timer.start();
 175       for (int k=0; k<REPEAT; ++k)
 176         mtlV2 = trans(m1) * mtlV1;
 177       timer.stop();
 178       std::cout << "   a' * v:\t" << timer.value() << endl;
 179     }
 180     #endif
 181
 182     std::cout << "\n\n";
 183   }
 184
 185   return 0;
 186 }
 187