1 // This file is part of Eigen, a lightweight C++ template library 2 // for linear algebra. 3 // 4 // Copyright (C) 2006-2008, 2010 Benoit Jacob <[email protected]> 5 // 6 // This Source Code Form is subject to the terms of the Mozilla 7 // Public License v. 2.0. If a copy of the MPL was not distributed 8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 10 #ifndef EIGEN_DOT_H 11 #define EIGEN_DOT_H 12 13 namespace Eigen { 14 15 namespace internal { 16 17 // helper function for dot(). The problem is that if we put that in the body of dot(), then upon calling dot 18 // with mismatched types, the compiler emits errors about failing to instantiate cwiseProduct BEFORE 19 // looking at the static assertions. Thus this is a trick to get better compile errors. 20 template<typename T, typename U, 21 // the NeedToTranspose condition here is taken straight from Assign.h 22 bool NeedToTranspose = T::IsVectorAtCompileTime 23 && U::IsVectorAtCompileTime 24 && ((int(T::RowsAtCompileTime) == 1 && int(U::ColsAtCompileTime) == 1) 25 | // FIXME | instead of || to please GCC 4.4.0 stupid warning "suggest parentheses around &&". 26 // revert to || as soon as not needed anymore. 27 (int(T::ColsAtCompileTime) == 1 && int(U::RowsAtCompileTime) == 1)) 28 > 29 struct dot_nocheck 30 { 31 typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod; 32 typedef typename conj_prod::result_type ResScalar; 33 EIGEN_DEVICE_FUNC 34 EIGEN_STRONG_INLINE rundot_nocheck35 static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b) 36 { 37 return a.template binaryExpr<conj_prod>(b).sum(); 38 } 39 }; 40 41 template<typename T, typename U> 42 struct dot_nocheck<T, U, true> 43 { 44 typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod; 45 typedef typename conj_prod::result_type ResScalar; 46 EIGEN_DEVICE_FUNC 47 EIGEN_STRONG_INLINE 48 static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b) 49 { 50 return a.transpose().template binaryExpr<conj_prod>(b).sum(); 51 } 52 }; 53 54 } // end namespace internal 55 56 /** \fn MatrixBase::dot 57 * \returns the dot product of *this with other. 58 * 59 * \only_for_vectors 60 * 61 * \note If the scalar type is complex numbers, then this function returns the hermitian 62 * (sesquilinear) dot product, conjugate-linear in the first variable and linear in the 63 * second variable. 64 * 65 * \sa squaredNorm(), norm() 66 */ 67 template<typename Derived> 68 template<typename OtherDerived> 69 EIGEN_DEVICE_FUNC 70 EIGEN_STRONG_INLINE 71 typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType 72 MatrixBase<Derived>::dot(const MatrixBase<OtherDerived>& other) const 73 { 74 EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) 75 EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived) 76 EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived) 77 #if !(defined(EIGEN_NO_STATIC_ASSERT) && defined(EIGEN_NO_DEBUG)) 78 typedef internal::scalar_conj_product_op<Scalar,typename OtherDerived::Scalar> func; 79 EIGEN_CHECK_BINARY_COMPATIBILIY(func,Scalar,typename OtherDerived::Scalar); 80 #endif 81 82 eigen_assert(size() == other.size()); 83 84 return internal::dot_nocheck<Derived,OtherDerived>::run(*this, other); 85 } 86 87 //---------- implementation of L2 norm and related functions ---------- 88 89 /** \returns, for vectors, the squared \em l2 norm of \c *this, and for matrices the squared Frobenius norm. 90 * In both cases, it consists in the sum of the square of all the matrix entries. 91 * For vectors, this is also equals to the dot product of \c *this with itself. 92 * 93 * \sa dot(), norm(), lpNorm() 94 */ 95 template<typename Derived> 96 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::squaredNorm() const 97 { 98 return numext::real((*this).cwiseAbs2().sum()); 99 } 100 101 /** \returns, for vectors, the \em l2 norm of \c *this, and for matrices the Frobenius norm. 102 * In both cases, it consists in the square root of the sum of the square of all the matrix entries. 103 * For vectors, this is also equals to the square root of the dot product of \c *this with itself. 104 * 105 * \sa lpNorm(), dot(), squaredNorm() 106 */ 107 template<typename Derived> 108 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const 109 { 110 return numext::sqrt(squaredNorm()); 111 } 112 113 /** \returns an expression of the quotient of \c *this by its own norm. 114 * 115 * \warning If the input vector is too small (i.e., this->norm()==0), 116 * then this function returns a copy of the input. 117 * 118 * \only_for_vectors 119 * 120 * \sa norm(), normalize() 121 */ 122 template<typename Derived> 123 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject 124 MatrixBase<Derived>::normalized() const 125 { 126 typedef typename internal::nested_eval<Derived,2>::type _Nested; 127 _Nested n(derived()); 128 RealScalar z = n.squaredNorm(); 129 // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU 130 if(z>RealScalar(0)) 131 return n / numext::sqrt(z); 132 else 133 return n; 134 } 135 136 /** Normalizes the vector, i.e. divides it by its own norm. 137 * 138 * \only_for_vectors 139 * 140 * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged. 141 * 142 * \sa norm(), normalized() 143 */ 144 template<typename Derived> 145 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void MatrixBase<Derived>::normalize() 146 { 147 RealScalar z = squaredNorm(); 148 // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU 149 if(z>RealScalar(0)) 150 derived() /= numext::sqrt(z); 151 } 152 153 /** \returns an expression of the quotient of \c *this by its own norm while avoiding underflow and overflow. 154 * 155 * \only_for_vectors 156 * 157 * This method is analogue to the normalized() method, but it reduces the risk of 158 * underflow and overflow when computing the norm. 159 * 160 * \warning If the input vector is too small (i.e., this->norm()==0), 161 * then this function returns a copy of the input. 162 * 163 * \sa stableNorm(), stableNormalize(), normalized() 164 */ 165 template<typename Derived> 166 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject 167 MatrixBase<Derived>::stableNormalized() const 168 { 169 typedef typename internal::nested_eval<Derived,3>::type _Nested; 170 _Nested n(derived()); 171 RealScalar w = n.cwiseAbs().maxCoeff(); 172 RealScalar z = (n/w).squaredNorm(); 173 if(z>RealScalar(0)) 174 return n / (numext::sqrt(z)*w); 175 else 176 return n; 177 } 178 179 /** Normalizes the vector while avoid underflow and overflow 180 * 181 * \only_for_vectors 182 * 183 * This method is analogue to the normalize() method, but it reduces the risk of 184 * underflow and overflow when computing the norm. 185 * 186 * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged. 187 * 188 * \sa stableNorm(), stableNormalized(), normalize() 189 */ 190 template<typename Derived> 191 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void MatrixBase<Derived>::stableNormalize() 192 { 193 RealScalar w = cwiseAbs().maxCoeff(); 194 RealScalar z = (derived()/w).squaredNorm(); 195 if(z>RealScalar(0)) 196 derived() /= numext::sqrt(z)*w; 197 } 198 199 //---------- implementation of other norms ---------- 200 201 namespace internal { 202 203 template<typename Derived, int p> 204 struct lpNorm_selector 205 { 206 typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar; 207 EIGEN_DEVICE_FUNC 208 static inline RealScalar run(const MatrixBase<Derived>& m) 209 { 210 EIGEN_USING_STD(pow) 211 return pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p); 212 } 213 }; 214 215 template<typename Derived> 216 struct lpNorm_selector<Derived, 1> 217 { 218 EIGEN_DEVICE_FUNC 219 static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m) 220 { 221 return m.cwiseAbs().sum(); 222 } 223 }; 224 225 template<typename Derived> 226 struct lpNorm_selector<Derived, 2> 227 { 228 EIGEN_DEVICE_FUNC 229 static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m) 230 { 231 return m.norm(); 232 } 233 }; 234 235 template<typename Derived> 236 struct lpNorm_selector<Derived, Infinity> 237 { 238 typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar; 239 EIGEN_DEVICE_FUNC 240 static inline RealScalar run(const MatrixBase<Derived>& m) 241 { 242 if(Derived::SizeAtCompileTime==0 || (Derived::SizeAtCompileTime==Dynamic && m.size()==0)) 243 return RealScalar(0); 244 return m.cwiseAbs().maxCoeff(); 245 } 246 }; 247 248 } // end namespace internal 249 250 /** \returns the \b coefficient-wise \f$ \ell^p \f$ norm of \c *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values 251 * of the coefficients of \c *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^\infty \f$ 252 * norm, that is the maximum of the absolute values of the coefficients of \c *this. 253 * 254 * In all cases, if \c *this is empty, then the value 0 is returned. 255 * 256 * \note For matrices, this function does not compute the <a href="https://en.wikipedia.org/wiki/Operator_norm">operator-norm</a>. That is, if \c *this is a matrix, then its coefficients are interpreted as a 1D vector. Nonetheless, you can easily compute the 1-norm and \f$\infty\f$-norm matrix operator norms using \link TutorialReductionsVisitorsBroadcastingReductionsNorm partial reductions \endlink. 257 * 258 * \sa norm() 259 */ 260 template<typename Derived> 261 template<int p> 262 #ifndef EIGEN_PARSED_BY_DOXYGEN 263 EIGEN_DEVICE_FUNC inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real 264 #else 265 EIGEN_DEVICE_FUNC MatrixBase<Derived>::RealScalar 266 #endif 267 MatrixBase<Derived>::lpNorm() const 268 { 269 return internal::lpNorm_selector<Derived, p>::run(*this); 270 } 271 272 //---------- implementation of isOrthogonal / isUnitary ---------- 273 274 /** \returns true if *this is approximately orthogonal to \a other, 275 * within the precision given by \a prec. 276 * 277 * Example: \include MatrixBase_isOrthogonal.cpp 278 * Output: \verbinclude MatrixBase_isOrthogonal.out 279 */ 280 template<typename Derived> 281 template<typename OtherDerived> 282 bool MatrixBase<Derived>::isOrthogonal 283 (const MatrixBase<OtherDerived>& other, const RealScalar& prec) const 284 { 285 typename internal::nested_eval<Derived,2>::type nested(derived()); 286 typename internal::nested_eval<OtherDerived,2>::type otherNested(other.derived()); 287 return numext::abs2(nested.dot(otherNested)) <= prec * prec * nested.squaredNorm() * otherNested.squaredNorm(); 288 } 289 290 /** \returns true if *this is approximately an unitary matrix, 291 * within the precision given by \a prec. In the case where the \a Scalar 292 * type is real numbers, a unitary matrix is an orthogonal matrix, whence the name. 293 * 294 * \note This can be used to check whether a family of vectors forms an orthonormal basis. 295 * Indeed, \c m.isUnitary() returns true if and only if the columns (equivalently, the rows) of m form an 296 * orthonormal basis. 297 * 298 * Example: \include MatrixBase_isUnitary.cpp 299 * Output: \verbinclude MatrixBase_isUnitary.out 300 */ 301 template<typename Derived> 302 bool MatrixBase<Derived>::isUnitary(const RealScalar& prec) const 303 { 304 typename internal::nested_eval<Derived,1>::type self(derived()); 305 for(Index i = 0; i < cols(); ++i) 306 { 307 if(!internal::isApprox(self.col(i).squaredNorm(), static_cast<RealScalar>(1), prec)) 308 return false; 309 for(Index j = 0; j < i; ++j) 310 if(!internal::isMuchSmallerThan(self.col(i).dot(self.col(j)), static_cast<Scalar>(1), prec)) 311 return false; 312 } 313 return true; 314 } 315 316 } // end namespace Eigen 317 318 #endif // EIGEN_DOT_H 319