1 // This file is part of Eigen, a lightweight C++ template library 2 // for linear algebra. 3 // 4 // Copyright (C) 2008 Gael Guennebaud <gael.guennebaud (at) inria.fr> 5 // Copyright (C) 2009 Benoit Jacob <jacob.benoit.1 (at) gmail.com> 6 // 7 // This Source Code Form is subject to the terms of the Mozilla 8 // Public License v. 2.0. If a copy of the MPL was not distributed 9 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 11 #include "svd_common.h" 12 13 template<typename MatrixType, int QRPreconditioner> 14 void jacobisvd_check_full(const MatrixType& m, const JacobiSVD<MatrixType, QRPreconditioner>& svd) 15 { 16 svd_check_full<MatrixType, JacobiSVD<MatrixType, QRPreconditioner > >(m, svd); 17 } 18 19 template<typename MatrixType, int QRPreconditioner> 20 void jacobisvd_compare_to_full(const MatrixType& m, 21 unsigned int computationOptions, 22 const JacobiSVD<MatrixType, QRPreconditioner>& referenceSvd) 23 { 24 svd_compare_to_full<MatrixType, JacobiSVD<MatrixType, QRPreconditioner> >(m, computationOptions, referenceSvd); 25 } 26 27 28 template<typename MatrixType, int QRPreconditioner> 29 void jacobisvd_solve(const MatrixType& m, unsigned int computationOptions) 30 { 31 svd_solve< MatrixType, JacobiSVD< MatrixType, QRPreconditioner > >(m, computationOptions); 32 } 33 34 35 36 template<typename MatrixType, int QRPreconditioner> 37 void jacobisvd_test_all_computation_options(const MatrixType& m) 38 { 39 40 if (QRPreconditioner == NoQRPreconditioner && m.rows() != m.cols()) 41 return; 42 43 JacobiSVD< MatrixType, QRPreconditioner > fullSvd(m, ComputeFullU|ComputeFullV); 44 svd_test_computation_options_1< MatrixType, JacobiSVD< MatrixType, QRPreconditioner > >(m, fullSvd); 45 46 if(QRPreconditioner == FullPivHouseholderQRPreconditioner) 47 return; 48 svd_test_computation_options_2< MatrixType, JacobiSVD< MatrixType, QRPreconditioner > >(m, fullSvd); 49 50 } 51 52 template<typename MatrixType> 53 void jacobisvd(const MatrixType& a = MatrixType(), bool pickrandom = true) 54 { 55 MatrixType m = pickrandom ? MatrixType::Random(a.rows(), a.cols()) : a; 56 57 jacobisvd_test_all_computation_options<MatrixType, FullPivHouseholderQRPreconditioner>(m); 58 jacobisvd_test_all_computation_options<MatrixType, ColPivHouseholderQRPreconditioner>(m); 59 jacobisvd_test_all_computation_options<MatrixType, HouseholderQRPreconditioner>(m); 60 jacobisvd_test_all_computation_options<MatrixType, NoQRPreconditioner>(m); 61 } 62 63 64 template<typename MatrixType> 65 void jacobisvd_verify_assert(const MatrixType& m) 66 { 67 68 svd_verify_assert<MatrixType, JacobiSVD< MatrixType > >(m); 69 70 typedef typename MatrixType::Index Index; 71 Index rows = m.rows(); 72 Index cols = m.cols(); 73 74 enum { 75 RowsAtCompileTime = MatrixType::RowsAtCompileTime, 76 ColsAtCompileTime = MatrixType::ColsAtCompileTime 77 }; 78 79 MatrixType a = MatrixType::Zero(rows, cols); 80 a.setZero(); 81 82 if (ColsAtCompileTime == Dynamic) 83 { 84 JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner> svd_fullqr; 85 VERIFY_RAISES_ASSERT(svd_fullqr.compute(a, ComputeFullU|ComputeThinV)) 86 VERIFY_RAISES_ASSERT(svd_fullqr.compute(a, ComputeThinU|ComputeThinV)) 87 VERIFY_RAISES_ASSERT(svd_fullqr.compute(a, ComputeThinU|ComputeFullV)) 88 } 89 } 90 91 template<typename MatrixType> 92 void jacobisvd_method() 93 { 94 enum { Size = MatrixType::RowsAtCompileTime }; 95 typedef typename MatrixType::RealScalar RealScalar; 96 typedef Matrix<RealScalar, Size, 1> RealVecType; 97 MatrixType m = MatrixType::Identity(); 98 VERIFY_IS_APPROX(m.jacobiSvd().singularValues(), RealVecType::Ones()); 99 VERIFY_RAISES_ASSERT(m.jacobiSvd().matrixU()); 100 VERIFY_RAISES_ASSERT(m.jacobiSvd().matrixV()); 101 VERIFY_IS_APPROX(m.jacobiSvd(ComputeFullU|ComputeFullV).solve(m), m); 102 } 103 104 105 106 template<typename MatrixType> 107 void jacobisvd_inf_nan() 108 { 109 svd_inf_nan<MatrixType, JacobiSVD< MatrixType > >(); 110 } 111 112 113 // Regression test for bug 286: JacobiSVD loops indefinitely with some 114 // matrices containing denormal numbers. 115 void jacobisvd_bug286() 116 { 117 #if defined __INTEL_COMPILER 118 // shut up warning #239: floating point underflow 119 #pragma warning push 120 #pragma warning disable 239 121 #endif 122 Matrix2d M; 123 M << -7.90884e-313, -4.94e-324, 124 0, 5.60844e-313; 125 #if defined __INTEL_COMPILER 126 #pragma warning pop 127 #endif 128 JacobiSVD<Matrix2d> svd; 129 svd.compute(M); // just check we don't loop indefinitely 130 } 131 132 133 void jacobisvd_preallocate() 134 { 135 svd_preallocate< JacobiSVD <MatrixXf> >(); 136 } 137 138 void test_jacobisvd() 139 { 140 CALL_SUBTEST_11(( jacobisvd<Matrix<double,Dynamic,Dynamic> > 141 (Matrix<double,Dynamic,Dynamic>(16, 6)) )); 142 143 CALL_SUBTEST_3(( jacobisvd_verify_assert(Matrix3f()) )); 144 CALL_SUBTEST_4(( jacobisvd_verify_assert(Matrix4d()) )); 145 CALL_SUBTEST_7(( jacobisvd_verify_assert(MatrixXf(10,12)) )); 146 CALL_SUBTEST_8(( jacobisvd_verify_assert(MatrixXcd(7,5)) )); 147 148 for(int i = 0; i < g_repeat; i++) { 149 Matrix2cd m; 150 m << 0, 1, 151 0, 1; 152 CALL_SUBTEST_1(( jacobisvd(m, false) )); 153 m << 1, 0, 154 1, 0; 155 CALL_SUBTEST_1(( jacobisvd(m, false) )); 156 157 Matrix2d n; 158 n << 0, 0, 159 0, 0; 160 CALL_SUBTEST_2(( jacobisvd(n, false) )); 161 n << 0, 0, 162 0, 1; 163 CALL_SUBTEST_2(( jacobisvd(n, false) )); 164 165 CALL_SUBTEST_3(( jacobisvd<Matrix3f>() )); 166 CALL_SUBTEST_4(( jacobisvd<Matrix4d>() )); 167 CALL_SUBTEST_5(( jacobisvd<Matrix<float,3,5> >() )); 168 CALL_SUBTEST_6(( jacobisvd<Matrix<double,Dynamic,2> >(Matrix<double,Dynamic,2>(10,2)) )); 169 170 int r = internal::random<int>(1, 30), 171 c = internal::random<int>(1, 30); 172 CALL_SUBTEST_7(( jacobisvd<MatrixXf>(MatrixXf(r,c)) )); 173 CALL_SUBTEST_8(( jacobisvd<MatrixXcd>(MatrixXcd(r,c)) )); 174 (void) r; 175 (void) c; 176 177 // Test on inf/nan matrix 178 CALL_SUBTEST_7( jacobisvd_inf_nan<MatrixXf>() ); 179 } 180 181 CALL_SUBTEST_7(( jacobisvd<MatrixXf>(MatrixXf(internal::random<int>(EIGEN_TEST_MAX_SIZE/4, EIGEN_TEST_MAX_SIZE/2), internal::random<int>(EIGEN_TEST_MAX_SIZE/4, EIGEN_TEST_MAX_SIZE/2))) )); 182 CALL_SUBTEST_8(( jacobisvd<MatrixXcd>(MatrixXcd(internal::random<int>(EIGEN_TEST_MAX_SIZE/4, EIGEN_TEST_MAX_SIZE/3), internal::random<int>(EIGEN_TEST_MAX_SIZE/4, EIGEN_TEST_MAX_SIZE/3))) )); 183 184 185 // test matrixbase method 186 CALL_SUBTEST_1(( jacobisvd_method<Matrix2cd>() )); 187 CALL_SUBTEST_3(( jacobisvd_method<Matrix3f>() )); 188 189 190 // Test problem size constructors 191 CALL_SUBTEST_7( JacobiSVD<MatrixXf>(10,10) ); 192 193 // Check that preallocation avoids subsequent mallocs 194 CALL_SUBTEST_9( jacobisvd_preallocate() ); 195 196 // Regression check for bug 286 197 CALL_SUBTEST_2( jacobisvd_bug286() ); 198 } 199
