# include "sandia_rules.hpp" # include "sgmga.hpp" # include # include # include # include # include int main ( ); void sgmga_vcn_tests ( ); void sgmga_vcn_test ( int dim_num, double importance[], double level_weight[], double q_min, double q_max ); void sgmga_vcn_timing_tests ( ); void sgmga_vcn_timing_test ( int dim_num, double importance[], double level_weight[], double q_min, double q_max ); void sgmga_vcn_ordered_tests ( ); void sgmga_vcn_ordered_test ( int dim_num, double importance[], double level_weight[], double q_min, double q_max ); //****************************************************************************80 int main ( ) //****************************************************************************80 // // Purpose: // // MAIN is the main program for SGMGA_VCN_PRB. // // Discussion: // // SGMGA_VCN_PRB tests SGMGA_VCN. // // Licensing: // // This code is distributed under the GNU LGPL license. // // Modified: // // 21 May 2010 // // Author: // // John Burkardt // { webbur::timestamp ( ); std::cout << "\n"; std::cout << "SGMGA_VCN_PRB\n"; std::cout << " C++ version\n"; std::cout << " Test the SGMGA_VCN and SGMGA_VCN_ORDERED functions.\n"; sgmga_vcn_tests ( ); sgmga_vcn_timing_tests ( ); sgmga_vcn_ordered_tests ( ); std::cout << "\n"; std::cout << "SGMGA_VCN_PRB\n"; std::cout << " Normal end of execution.\n"; std::cout << "\n"; webbur::timestamp ( ); return 0; } //****************************************************************************80 void sgmga_vcn_tests ( ) //****************************************************************************80 // // Purpose: // // SGMGA_VCN_TESTS calls SGMGA_VCN_TEST. // // Licensing: // // This code is distributed under the GNU LGPL license. // // Modified: // // 21 May 2010 // // Author: // // John Burkardt // { int dim; int dim_num; int dim_num_array[12] = { 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 4, 4 }; double *importance; int level_max; int level_max_array[12] = { 0, 1, 2, 3, 4, 0, 1, 2, 3, 4, 2, 3 }; double *level_weight; double q_max; double q_min; int test; int test_num = 12; std::cout << "\n"; std::cout << "SGMGA_VCN_TESTS\n"; std::cout << " calls SGMGA_VCN_TEST.\n"; // // Isotropic examples. // for ( test = 0; test < test_num; test++ ) { dim_num = dim_num_array[test]; importance = new double[dim_num]; for ( dim = 0; dim < dim_num; dim++ ) { importance[dim] = 1.0; } level_weight = new double[dim_num]; webbur::sgmga_importance_to_aniso ( dim_num, importance, level_weight ); level_max = level_max_array[test]; q_min = ( double ) ( level_max ) - webbur::r8vec_sum ( dim_num, level_weight ); q_max = ( double ) ( level_max ); sgmga_vcn_test ( dim_num, importance, level_weight, q_min, q_max ); delete [] importance; delete [] level_weight; } // // Zero weight example. // dim_num = 3; importance = new double[dim_num]; importance[0] = 1.0; importance[1] = 0.0; importance[2] = 1.0; level_weight = new double[dim_num]; webbur::sgmga_importance_to_aniso ( dim_num, importance, level_weight ); level_max = 2; q_min = ( double ) ( level_max ) - webbur::r8vec_sum ( dim_num, level_weight ); q_max = ( double ) ( level_max ); sgmga_vcn_test ( dim_num, importance, level_weight, q_min, q_max ); delete [] importance; delete [] level_weight; // // Anisotropic examples. // for ( test = 0; test < test_num; test++ ) { dim_num = dim_num_array[test]; importance = new double[dim_num]; for ( dim = 0; dim < dim_num; dim++ ) { importance[dim] = ( double ) ( dim + 1 ); } level_weight = new double[dim_num]; webbur::sgmga_importance_to_aniso ( dim_num, importance, level_weight ); level_max = level_max_array[test]; q_min = ( double ) ( level_max ) - webbur::r8vec_sum ( dim_num, level_weight ); q_max = ( double ) ( level_max ); sgmga_vcn_test ( dim_num, importance, level_weight, q_min, q_max ); delete [] importance; delete [] level_weight; } return; } //****************************************************************************80 void sgmga_vcn_test ( int dim_num, double importance[], double level_weight[], double q_min, double q_max ) //****************************************************************************80 // // Purpose: // // SGMGA_VCN_TEST tests SGMGA_VCN. // // Licensing: // // This code is distributed under the GNU LGPL license. // // Modified: // // 26 April 2011 // // Author: // // John Burkardt // { int dim; int i; int *level_1d; int *level_1d_max; int *level_1d_min; double level_weight_norm; bool more_grids; double q; int test; level_1d = new int[dim_num]; level_1d_max = new int[dim_num]; level_1d_min = new int[dim_num]; std::cout << "\n"; std::cout << "SGMGA_VCN_TEST\n"; std::cout << " Consider vectors 0 <= LEVEL_1D(1:N) <= LEVEL_1D_MAX(1:N),\n"; std::cout << " Set Q = sum ( LEVEL_WEIGHT(1:N) * LEVEL_1D(1:N) )\n"; std::cout << " Accept vectors for which Q_MIN < Q <= Q_MAX\n"; std::cout << " No particular order is imposed on the LEVEL_1D values.\n"; std::cout << " SGMGA_VCN_NAIVE uses a naive approach;\n"; std::cout << " SGMGA_VCN tries to be more efficient.\n"; std::cout << " Here, we just compare the results.\n"; for ( dim = 0; dim < dim_num; dim++ ) { if ( 0.0 < level_weight[dim] ) { level_1d_max[dim] = webbur::r8_floor ( q_max / level_weight[dim] ) + 1; if ( q_max <= ( level_1d_max[dim] - 1 ) * level_weight[dim] ) { level_1d_max[dim] = level_1d_max[dim] - 1; } } else { level_1d_max[dim] = 0; } } for ( dim = 0; dim < dim_num; dim++ ) { level_1d_min[dim] = 0; } more_grids = false; std::cout << "\n"; std::cout << " IMPORTANCE:\n"; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(14) << importance[dim]; } std::cout << "\n"; std::cout << " LEVEL_WEIGHT:\n"; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(14) << level_weight[dim]; } std::cout << "\n"; std::cout << "\n"; std::cout << " SGMGA_VCN_NAIVE\n"; std::cout << " I Q X\n"; std::cout << " MIN" << " " << std::setw(14) << q_min; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_min[dim]; } std::cout << "\n"; i = 0; for ( ; ; ) { webbur::sgmga_vcn_naive ( dim_num, level_weight, level_1d_max, level_1d, q_min, q_max, &more_grids ); if ( !more_grids ) { break; } q = 0.0; for ( dim = 0; dim < dim_num; dim++ ) { q = q + level_weight[dim] * ( double ) level_1d[dim]; } i = i + 1; std::cout << " " << std::setw(4) << i << " " << std::setw(14) << q; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d[dim]; } std::cout << "\n"; } std::cout << " MAX" << " " << std::setw(14) << q_max; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_max[dim]; } std::cout << "\n"; std::cout << "\n"; std::cout << " SGMGA_VCN\n"; std::cout << " I Q X\n"; std::cout << " MIN" << " " << std::setw(14) << q_min; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_min[dim]; } std::cout << "\n"; i = 0; for ( ; ; ) { webbur::sgmga_vcn ( dim_num, level_weight,level_1d, q_min, q_max, &more_grids ); if ( !more_grids ) { break; } q = 0.0; for ( dim = 0; dim < dim_num; dim++ ) { q = q + level_weight[dim] * ( double ) level_1d[dim]; } i = i + 1; std::cout << " " << std::setw(4) << i << " " << std::setw(14) << q; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d[dim]; } std::cout << "\n"; } std::cout << " MAX" << " " << std::setw(14) << q_max; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_max[dim]; } std::cout << "\n"; delete [] level_1d; delete [] level_1d_max; delete [] level_1d_min; return; } //****************************************************************************80 void sgmga_vcn_timing_tests ( ) //****************************************************************************80 // // Purpose: // // SGMGA_VCN_TIMING_TESTS calls SGMGA_VCN_TIMING_TEST. // // Licensing: // // This code is distributed under the GNU LGPL license. // // Modified: // // 21 May 2010 // // Author: // // John Burkardt // { int dim; int dim_num; double *importance; int level_max; double *level_weight; double q_max; double q_min; int test; std::cout << "\n"; std::cout << "SGMGA_VCN_TIMING_TESTS\n"; std::cout << " calls SGMGA_VCN_TIMING_TEST.\n"; // // Isotropic examples. // dim_num = 2; for ( test = 0; test < 2; test++ ) { dim_num = dim_num * 2; importance = new double[dim_num]; for ( dim = 0; dim < dim_num; dim++ ) { importance[dim] = 1.0; } level_weight = new double[dim_num]; webbur::sgmga_importance_to_aniso ( dim_num, importance, level_weight ); level_max = 2; q_min = ( double ) ( level_max ) - webbur::r8vec_sum ( dim_num, level_weight ); q_max = ( double ) ( level_max ); sgmga_vcn_timing_test ( dim_num, importance, level_weight, q_min, q_max ); delete [] importance; delete [] level_weight; } // // Anisotropic examples. // dim_num = 2; for ( test = 0; test < 2; test++ ) { dim_num = dim_num * 2; importance = new double[dim_num]; for ( dim = 0; dim < dim_num; dim++ ) { importance[dim] = ( double ) ( dim + 1 ); } level_weight = new double[dim_num]; webbur::sgmga_importance_to_aniso ( dim_num, importance, level_weight ); level_max = 2; q_min = ( double ) ( level_max ) - webbur::r8vec_sum ( dim_num, level_weight ); q_max = ( double ) ( level_max ); sgmga_vcn_timing_test ( dim_num, importance, level_weight, q_min, q_max ); delete [] importance; delete [] level_weight; } return; } //****************************************************************************80 void sgmga_vcn_timing_test ( int dim_num, double importance[], double level_weight[], double q_min, double q_max ) //****************************************************************************80 // // Purpose: // // SGMGA_VCN_TIMING_TEST times SGMGA_VCN and SGMGA_VCN_NAIVE. // // Licensing: // // This code is distributed under the GNU LGPL license. // // Modified: // // 26 April 2011 // // Author: // // John Burkardt // { int dim; int i; int *level_1d; int *level_1d_max; int *level_1d_min; double level_weight_norm; bool more_grids; double q; clock_t t1; clock_t t2; int test; level_1d = new int[dim_num]; level_1d_max = new int[dim_num]; level_1d_min = new int[dim_num]; std::cout << "\n"; std::cout << "SGMGA_VCN_TIMING_TEST\n"; std::cout << " Consider vectors 0 <= LEVEL_1D(1:N) <= LEVEL_1D_MAX(1:N),\n"; std::cout << " Set Q = sum ( LEVEL_WEIGHT(1:N) * LEVEL_1D(1:N) )\n"; std::cout << " Accept vectors for which Q_MIN < Q <= Q_MAX\n"; std::cout << " No particular order is imposed on the LEVEL_1D values.\n"; std::cout << " SGMGA_VCN_NAIVE uses a naive approach;\n"; std::cout << " SGMGA_VCN tries to be more efficient.\n"; std::cout << " Here, we compare the timings.\n"; for ( dim = 0; dim < dim_num; dim++ ) { if ( 0.0 < level_weight[dim] ) { level_1d_max[dim] = webbur::r8_floor ( q_max / level_weight[dim] ) + 1; if ( q_max <= ( level_1d_max[dim] - 1 ) * level_weight[dim] ) { level_1d_max[dim] = level_1d_max[dim] - 1; } } else { level_1d_max[dim] = 0; } } for ( dim = 0; dim < dim_num; dim++ ) { level_1d_min[dim] = 0; } more_grids = false; std::cout << "\n"; std::cout << " IMPORTANCE:\n"; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(14) << importance[dim]; } std::cout << "\n"; std::cout << " LEVEL_WEIGHT:\n"; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(14) << level_weight[dim]; } std::cout << "\n"; std::cout << "\n"; std::cout << " SGMGA_VCN_NAIVE\n"; std::cout << " I Q X\n"; std::cout << " MIN" << " " << std::setw(14) << q_min; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_min[dim]; } std::cout << "\n"; i = 0; t1 = std::clock ( ); for ( ; ; ) { webbur::sgmga_vcn_naive ( dim_num, level_weight, level_1d_max, level_1d, q_min, q_max, &more_grids ); if ( !more_grids ) { break; } } t2 = std::clock ( ); std::cout << " MAX" << " " << std::setw(14) << q_max; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_max[dim]; } std::cout << "\n"; std::cout << " TIME" << " " << std::setw(14) << ( double ) ( t2 - t1 ) / CLOCKS_PER_SEC << "\n"; std::cout << "\n"; std::cout << " SGMGA_VCN\n"; std::cout << " I Q X\n"; std::cout << " MIN" << " " << std::setw(14) << q_min; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_min[dim]; } std::cout << "\n"; i = 0; t1 = std::clock ( ); for ( ; ; ) { webbur::sgmga_vcn ( dim_num, level_weight,level_1d, q_min, q_max, &more_grids ); if ( !more_grids ) { break; } } t2 = std::clock ( ); std::cout << " MAX" << " " << std::setw(14) << q_max; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_max[dim]; } std::cout << "\n"; std::cout << " TIME" << " " << std::setw(14) << ( double ) ( t2 - t1 ) / CLOCKS_PER_SEC << "\n"; delete [] level_1d; delete [] level_1d_max; delete [] level_1d_min; return; } //****************************************************************************80 void sgmga_vcn_ordered_tests ( ) //****************************************************************************80 // // Purpose: // // SGMGA_VCN_ORDERED_TESTS calls SGMGA_VCN_ORDERED_TEST. // // Licensing: // // This code is distributed under the GNU LGPL license. // // Modified: // // 21 May 2010 // // Author: // // John Burkardt // { int dim; int dim_num; int dim_num_array[12] = { 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 4, 4 }; double *importance; int level_max; int level_max_array[12] = { 0, 1, 2, 3, 4, 0, 1, 2, 3, 4, 2, 3 }; double *level_weight; double q_max; double q_min; int test; int test_num = 12; std::cout << "\n"; std::cout << "SGMGA_VCN_ORDERED_TESTS\n"; std::cout << " calls SGMGA_VCN_ORDERED_TEST.\n"; // // Isotropic examples. // for ( test = 0; test < test_num; test++ ) { dim_num = dim_num_array[test]; importance = new double[dim_num]; for ( dim = 0; dim < dim_num; dim++ ) { importance[dim] = 1.0; } level_weight = new double[dim_num]; webbur::sgmga_importance_to_aniso ( dim_num, importance, level_weight ); level_max = level_max_array[test]; q_min = ( double ) ( level_max ) - webbur::r8vec_sum ( dim_num, level_weight ); q_max = ( double ) ( level_max ); sgmga_vcn_ordered_test ( dim_num, importance, level_weight, q_min, q_max ); delete [] importance; delete [] level_weight; } // // Anisotropic examples. // for ( test = 0; test < test_num; test++ ) { dim_num = dim_num_array[test]; importance = new double[dim_num]; for ( dim = 0; dim < dim_num; dim++ ) { importance[dim] = ( double ) ( dim + 1 ); } level_weight = new double[dim_num]; webbur::sgmga_importance_to_aniso ( dim_num, importance, level_weight ); level_max = level_max_array[test]; q_min = ( double ) ( level_max ) - webbur::r8vec_sum ( dim_num, level_weight ); q_max = ( double ) ( level_max ); sgmga_vcn_ordered_test ( dim_num, importance, level_weight, q_min, q_max ); delete [] importance; delete [] level_weight; } return; } //****************************************************************************80 void sgmga_vcn_ordered_test ( int dim_num, double importance[], double level_weight[], double q_min, double q_max ) //****************************************************************************80 // // Purpose: // // SGMGA_VCN_ORDERED_TEST tests SGMGA_VCN_ORDERED and SGMGA_VCN_ORDERED_NAIVE. // // Licensing: // // This code is distributed under the GNU LGPL license. // // Modified: // // 21 May 2010 // // Author: // // John Burkardt // { int dim; int i; int *level_1d; int *level_1d_max; int *level_1d_min; double level_weight_norm; bool more_grids; double q; int test; level_1d = new int[dim_num]; level_1d_max = new int[dim_num]; level_1d_min = new int[dim_num]; std::cout << "\n"; std::cout << "SGMGA_VCN_ORDERED_TEST\n"; std::cout << " Consider vectors 0 <= LEVEL_1D(1:N) <= LEVEL_1D_MAX(1:N),\n"; std::cout << " Set Q = sum ( LEVEL_WEIGHT(1:N) * LEVEL_1D(1:N) )\n"; std::cout << " Accept only vectors for which Q_MIN < Q <= Q_MAX\n"; std::cout << " The solutions are weakly ordered by the value of Q.\n"; std::cout << " SGMGA_VCN_ORDERED_NAIVE calls SGMGA_VCN_NAIVE;\n"; std::cout << " SGMGA_VCN_ORDERED calls SGMGA_VCN.\n"; for ( dim = 0; dim < dim_num; dim++ ) { if ( 0.0 < level_weight[dim] ) { level_1d_max[dim] = webbur::r8_floor ( q_max / level_weight[dim] ) + 1; } else { level_1d_max[dim] = 0; } } for ( dim = 0; dim < dim_num; dim++ ) { level_1d_min[dim] = 0; } more_grids = false; std::cout << "\n"; std::cout << " IMPORTANCE:\n"; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(14) << importance[dim]; } std::cout << "\n"; std::cout << " LEVEL_WEIGHT:\n"; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(14) << level_weight[dim]; } std::cout << "\n"; std::cout << "\n"; std::cout << " SGMGA_VCN_ORDERED_NAIVE:\n"; std::cout << " I Q X\n"; std::cout << " MIN" << " " << std::setw(14) << q_min; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_min[dim]; } std::cout << "\n"; i = 0; for ( ; ; ) { webbur::sgmga_vcn_ordered_naive ( dim_num, level_weight, level_1d_max, level_1d, q_min, q_max, &more_grids ); if ( !more_grids ) { break; } q = 0.0; for ( dim = 0; dim < dim_num; dim++ ) { q = q + level_weight[dim] * ( double ) level_1d[dim]; } i = i + 1; std::cout << " " << std::setw(4) << i << " " << std::setw(14) << q; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d[dim]; } std::cout << "\n"; } std::cout << " MAX" << " " << std::setw(14) << q_max; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_max[dim]; } std::cout << "\n"; std::cout << "\n"; std::cout << " SGMGA_VCN_ORDERED:\n"; std::cout << " I Q X\n"; std::cout << " MIN" << " " << std::setw(14) << q_min; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_min[dim]; } std::cout << "\n"; i = 0; for ( ; ; ) { webbur::sgmga_vcn_ordered ( dim_num, level_weight, level_1d_max, level_1d, q_min, q_max, &more_grids ); if ( !more_grids ) { break; } q = 0.0; for ( dim = 0; dim < dim_num; dim++ ) { q = q + level_weight[dim] * ( double ) level_1d[dim]; } i = i + 1; std::cout << " " << std::setw(4) << i << " " << std::setw(14) << q; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d[dim]; } std::cout << "\n"; } std::cout << " MAX" << " " << std::setw(14) << q_max; for ( dim = 0; dim < dim_num; dim++ ) { std::cout << " " << std::setw(2) << level_1d_max[dim]; } std::cout << "\n"; delete [] level_1d; delete [] level_1d_max; delete [] level_1d_min; return; }