//g++ -O3 -g0 -DNDEBUG sparse_product.cpp -I.. -I/home/gael/Coding/LinearAlgebra/mtl4/ -DDENSITY=0.005 -DSIZE=10000 && ./a.out //g++ -O3 -g0 -DNDEBUG sparse_product.cpp -I.. -I/home/gael/Coding/LinearAlgebra/mtl4/ -DDENSITY=0.05 -DSIZE=2000 && ./a.out // -DNOGMM -DNOMTL -DCSPARSE // -I /home/gael/Coding/LinearAlgebra/CSparse/Include/ /home/gael/Coding/LinearAlgebra/CSparse/Lib/libcsparse.a #ifndef SIZE #define SIZE 100000 #endif #ifndef NBPERROW #define NBPERROW 24 #endif #ifndef REPEAT #define REPEAT 2 #endif #ifndef NBTRIES #define NBTRIES 2 #endif #ifndef KK #define KK 10 #endif #ifndef NOGOOGLE #define EIGEN_GOOGLEHASH_SUPPORT #include <google/sparse_hash_map> #endif #include "BenchSparseUtil.h" #define CHECK_MEM // #define CHECK_MEM std/**/::cout << "check mem\n"; getchar(); #define BENCH(X) \ timer.reset(); \ for (int _j=0; _j<NBTRIES; ++_j) { \ timer.start(); \ for (int _k=0; _k<REPEAT; ++_k) { \ X \ } timer.stop(); } typedef std::vector<Vector2i> Coordinates; typedef std::vector<float> Values; EIGEN_DONT_INLINE Scalar* setinnerrand_eigen(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_eigen_dynamic(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_eigen_compact(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_eigen_sumeq(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_eigen_gnu_hash(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_eigen_google_dense(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_eigen_google_sparse(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_scipy(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_ublas_mapped(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_ublas_coord(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_ublas_compressed(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_ublas_genvec(const Coordinates& coords, const Values& vals); EIGEN_DONT_INLINE Scalar* setrand_mtl(const Coordinates& coords, const Values& vals); int main(int argc, char *argv[]) { int rows = SIZE; int cols = SIZE; bool fullyrand = true; BenchTimer timer; Coordinates coords; Values values; if(fullyrand) { Coordinates pool; pool.reserve(cols*NBPERROW); std::cerr << "fill pool" << "\n"; for (int i=0; i<cols*NBPERROW; ) { // DynamicSparseMatrix<int> stencil(SIZE,SIZE); Vector2i ij(internal::random<int>(0,rows-1),internal::random<int>(0,cols-1)); // if(stencil.coeffRef(ij.x(), ij.y())==0) { // stencil.coeffRef(ij.x(), ij.y()) = 1; pool.push_back(ij); } ++i; } std::cerr << "pool ok" << "\n"; int n = cols*NBPERROW*KK; coords.reserve(n); values.reserve(n); for (int i=0; i<n; ++i) { int i = internal::random<int>(0,pool.size()); coords.push_back(pool[i]); values.push_back(internal::random<Scalar>()); } } else { for (int j=0; j<cols; ++j) for (int i=0; i<NBPERROW; ++i) { coords.push_back(Vector2i(internal::random<int>(0,rows-1),j)); values.push_back(internal::random<Scalar>()); } } std::cout << "nnz = " << coords.size() << "\n"; CHECK_MEM // dense matrices #ifdef DENSEMATRIX { BENCH(setrand_eigen_dense(coords,values);) std::cout << "Eigen Dense\t" << timer.value() << "\n"; } #endif // eigen sparse matrices // if (!fullyrand) // { // BENCH(setinnerrand_eigen(coords,values);) // std::cout << "Eigen fillrand\t" << timer.value() << "\n"; // } { BENCH(setrand_eigen_dynamic(coords,values);) std::cout << "Eigen dynamic\t" << timer.value() << "\n"; } // { // BENCH(setrand_eigen_compact(coords,values);) // std::cout << "Eigen compact\t" << timer.value() << "\n"; // } { BENCH(setrand_eigen_sumeq(coords,values);) std::cout << "Eigen sumeq\t" << timer.value() << "\n"; } { // BENCH(setrand_eigen_gnu_hash(coords,values);) // std::cout << "Eigen std::map\t" << timer.value() << "\n"; } { BENCH(setrand_scipy(coords,values);) std::cout << "scipy\t" << timer.value() << "\n"; } #ifndef NOGOOGLE { BENCH(setrand_eigen_google_dense(coords,values);) std::cout << "Eigen google dense\t" << timer.value() << "\n"; } { BENCH(setrand_eigen_google_sparse(coords,values);) std::cout << "Eigen google sparse\t" << timer.value() << "\n"; } #endif #ifndef NOUBLAS { // BENCH(setrand_ublas_mapped(coords,values);) // std::cout << "ublas mapped\t" << timer.value() << "\n"; } { BENCH(setrand_ublas_genvec(coords,values);) std::cout << "ublas vecofvec\t" << timer.value() << "\n"; } /*{ timer.reset(); timer.start(); for (int k=0; k<REPEAT; ++k) setrand_ublas_compressed(coords,values); timer.stop(); std::cout << "ublas comp\t" << timer.value() << "\n"; } { timer.reset(); timer.start(); for (int k=0; k<REPEAT; ++k) setrand_ublas_coord(coords,values); timer.stop(); std::cout << "ublas coord\t" << timer.value() << "\n"; }*/ #endif // MTL4 #ifndef NOMTL { BENCH(setrand_mtl(coords,values)); std::cout << "MTL\t" << timer.value() << "\n"; } #endif return 0; } EIGEN_DONT_INLINE Scalar* setinnerrand_eigen(const Coordinates& coords, const Values& vals) { using namespace Eigen; SparseMatrix<Scalar> mat(SIZE,SIZE); //mat.startFill(2000000/*coords.size()*/); for (int i=0; i<coords.size(); ++i) { mat.insert(coords[i].x(), coords[i].y()) = vals[i]; } mat.finalize(); CHECK_MEM; return 0; } EIGEN_DONT_INLINE Scalar* setrand_eigen_dynamic(const Coordinates& coords, const Values& vals) { using namespace Eigen; DynamicSparseMatrix<Scalar> mat(SIZE,SIZE); mat.reserve(coords.size()/10); for (int i=0; i<coords.size(); ++i) { mat.coeffRef(coords[i].x(), coords[i].y()) += vals[i]; } mat.finalize(); CHECK_MEM; return &mat.coeffRef(coords[0].x(), coords[0].y()); } EIGEN_DONT_INLINE Scalar* setrand_eigen_sumeq(const Coordinates& coords, const Values& vals) { using namespace Eigen; int n = coords.size()/KK; DynamicSparseMatrix<Scalar> mat(SIZE,SIZE); for (int j=0; j<KK; ++j) { DynamicSparseMatrix<Scalar> aux(SIZE,SIZE); mat.reserve(n); for (int i=j*n; i<(j+1)*n; ++i) { aux.insert(coords[i].x(), coords[i].y()) += vals[i]; } aux.finalize(); mat += aux; } return &mat.coeffRef(coords[0].x(), coords[0].y()); } EIGEN_DONT_INLINE Scalar* setrand_eigen_compact(const Coordinates& coords, const Values& vals) { using namespace Eigen; DynamicSparseMatrix<Scalar> setter(SIZE,SIZE); setter.reserve(coords.size()/10); for (int i=0; i<coords.size(); ++i) { setter.coeffRef(coords[i].x(), coords[i].y()) += vals[i]; } SparseMatrix<Scalar> mat = setter; CHECK_MEM; return &mat.coeffRef(coords[0].x(), coords[0].y()); } EIGEN_DONT_INLINE Scalar* setrand_eigen_gnu_hash(const Coordinates& coords, const Values& vals) { using namespace Eigen; SparseMatrix<Scalar> mat(SIZE,SIZE); { RandomSetter<SparseMatrix<Scalar>, StdMapTraits > setter(mat); for (int i=0; i<coords.size(); ++i) { setter(coords[i].x(), coords[i].y()) += vals[i]; } CHECK_MEM; } return &mat.coeffRef(coords[0].x(), coords[0].y()); } #ifndef NOGOOGLE EIGEN_DONT_INLINE Scalar* setrand_eigen_google_dense(const Coordinates& coords, const Values& vals) { using namespace Eigen; SparseMatrix<Scalar> mat(SIZE,SIZE); { RandomSetter<SparseMatrix<Scalar>, GoogleDenseHashMapTraits> setter(mat); for (int i=0; i<coords.size(); ++i) setter(coords[i].x(), coords[i].y()) += vals[i]; CHECK_MEM; } return &mat.coeffRef(coords[0].x(), coords[0].y()); } EIGEN_DONT_INLINE Scalar* setrand_eigen_google_sparse(const Coordinates& coords, const Values& vals) { using namespace Eigen; SparseMatrix<Scalar> mat(SIZE,SIZE); { RandomSetter<SparseMatrix<Scalar>, GoogleSparseHashMapTraits> setter(mat); for (int i=0; i<coords.size(); ++i) setter(coords[i].x(), coords[i].y()) += vals[i]; CHECK_MEM; } return &mat.coeffRef(coords[0].x(), coords[0].y()); } #endif template <class T> void coo_tocsr(const int n_row, const int n_col, const int nnz, const Coordinates Aij, const Values Ax, int Bp[], int Bj[], T Bx[]) { //compute number of non-zero entries per row of A coo_tocsr std::fill(Bp, Bp + n_row, 0); for (int n = 0; n < nnz; n++){ Bp[Aij[n].x()]++; } //cumsum the nnz per row to get Bp[] for(int i = 0, cumsum = 0; i < n_row; i++){ int temp = Bp[i]; Bp[i] = cumsum; cumsum += temp; } Bp[n_row] = nnz; //write Aj,Ax into Bj,Bx for(int n = 0; n < nnz; n++){ int row = Aij[n].x(); int dest = Bp[row]; Bj[dest] = Aij[n].y(); Bx[dest] = Ax[n]; Bp[row]++; } for(int i = 0, last = 0; i <= n_row; i++){ int temp = Bp[i]; Bp[i] = last; last = temp; } //now Bp,Bj,Bx form a CSR representation (with possible duplicates) } template< class T1, class T2 > bool kv_pair_less(const std::pair<T1,T2>& x, const std::pair<T1,T2>& y){ return x.first < y.first; } template<class I, class T> void csr_sort_indices(const I n_row, const I Ap[], I Aj[], T Ax[]) { std::vector< std::pair<I,T> > temp; for(I i = 0; i < n_row; i++){ I row_start = Ap[i]; I row_end = Ap[i+1]; temp.clear(); for(I jj = row_start; jj < row_end; jj++){ temp.push_back(std::make_pair(Aj[jj],Ax[jj])); } std::sort(temp.begin(),temp.end(),kv_pair_less<I,T>); for(I jj = row_start, n = 0; jj < row_end; jj++, n++){ Aj[jj] = temp[n].first; Ax[jj] = temp[n].second; } } } template <class I, class T> void csr_sum_duplicates(const I n_row, const I n_col, I Ap[], I Aj[], T Ax[]) { I nnz = 0; I row_end = 0; for(I i = 0; i < n_row; i++){ I jj = row_end; row_end = Ap[i+1]; while( jj < row_end ){ I j = Aj[jj]; T x = Ax[jj]; jj++; while( jj < row_end && Aj[jj] == j ){ x += Ax[jj]; jj++; } Aj[nnz] = j; Ax[nnz] = x; nnz++; } Ap[i+1] = nnz; } } EIGEN_DONT_INLINE Scalar* setrand_scipy(const Coordinates& coords, const Values& vals) { using namespace Eigen; SparseMatrix<Scalar> mat(SIZE,SIZE); mat.resizeNonZeros(coords.size()); // std::cerr << "setrand_scipy...\n"; coo_tocsr<Scalar>(SIZE,SIZE, coords.size(), coords, vals, mat._outerIndexPtr(), mat._innerIndexPtr(), mat._valuePtr()); // std::cerr << "coo_tocsr ok\n"; csr_sort_indices(SIZE, mat._outerIndexPtr(), mat._innerIndexPtr(), mat._valuePtr()); csr_sum_duplicates(SIZE, SIZE, mat._outerIndexPtr(), mat._innerIndexPtr(), mat._valuePtr()); mat.resizeNonZeros(mat._outerIndexPtr()[SIZE]); return &mat.coeffRef(coords[0].x(), coords[0].y()); } #ifndef NOUBLAS EIGEN_DONT_INLINE Scalar* setrand_ublas_mapped(const Coordinates& coords, const Values& vals) { using namespace boost; using namespace boost::numeric; using namespace boost::numeric::ublas; mapped_matrix<Scalar> aux(SIZE,SIZE); for (int i=0; i<coords.size(); ++i) { aux(coords[i].x(), coords[i].y()) += vals[i]; } CHECK_MEM; compressed_matrix<Scalar> mat(aux); return 0;// &mat(coords[0].x(), coords[0].y()); } /*EIGEN_DONT_INLINE Scalar* setrand_ublas_coord(const Coordinates& coords, const Values& vals) { using namespace boost; using namespace boost::numeric; using namespace boost::numeric::ublas; coordinate_matrix<Scalar> aux(SIZE,SIZE); for (int i=0; i<coords.size(); ++i) { aux(coords[i].x(), coords[i].y()) = vals[i]; } compressed_matrix<Scalar> mat(aux); return 0;//&mat(coords[0].x(), coords[0].y()); } EIGEN_DONT_INLINE Scalar* setrand_ublas_compressed(const Coordinates& coords, const Values& vals) { using namespace boost; using namespace boost::numeric; using namespace boost::numeric::ublas; compressed_matrix<Scalar> mat(SIZE,SIZE); for (int i=0; i<coords.size(); ++i) { mat(coords[i].x(), coords[i].y()) = vals[i]; } return 0;//&mat(coords[0].x(), coords[0].y()); }*/ EIGEN_DONT_INLINE Scalar* setrand_ublas_genvec(const Coordinates& coords, const Values& vals) { using namespace boost; using namespace boost::numeric; using namespace boost::numeric::ublas; // ublas::vector<coordinate_vector<Scalar> > foo; generalized_vector_of_vector<Scalar, row_major, ublas::vector<coordinate_vector<Scalar> > > aux(SIZE,SIZE); for (int i=0; i<coords.size(); ++i) { aux(coords[i].x(), coords[i].y()) += vals[i]; } CHECK_MEM; compressed_matrix<Scalar,row_major> mat(aux); return 0;//&mat(coords[0].x(), coords[0].y()); } #endif #ifndef NOMTL EIGEN_DONT_INLINE void setrand_mtl(const Coordinates& coords, const Values& vals); #endif