• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
5 // Copyright (C) 2010 Daniel Lowengrub <lowdanie@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 #ifndef EIGEN_SPARSEVIEW_H
12 #define EIGEN_SPARSEVIEW_H
13 
14 namespace Eigen {
15 
16 namespace internal {
17 
18 template<typename MatrixType>
19 struct traits<SparseView<MatrixType> > : traits<MatrixType>
20 {
21   typedef typename MatrixType::StorageIndex StorageIndex;
22   typedef Sparse StorageKind;
23   enum {
24     Flags = int(traits<MatrixType>::Flags) & (RowMajorBit)
25   };
26 };
27 
28 } // end namespace internal
29 
30 /** \ingroup SparseCore_Module
31   * \class SparseView
32   *
33   * \brief Expression of a dense or sparse matrix with zero or too small values removed
34   *
35   * \tparam MatrixType the type of the object of which we are removing the small entries
36   *
37   * This class represents an expression of a given dense or sparse matrix with
38   * entries smaller than \c reference * \c epsilon are removed.
39   * It is the return type of MatrixBase::sparseView() and SparseMatrixBase::pruned()
40   * and most of the time this is the only way it is used.
41   *
42   * \sa MatrixBase::sparseView(), SparseMatrixBase::pruned()
43   */
44 template<typename MatrixType>
45 class SparseView : public SparseMatrixBase<SparseView<MatrixType> >
46 {
47   typedef typename MatrixType::Nested MatrixTypeNested;
48   typedef typename internal::remove_all<MatrixTypeNested>::type _MatrixTypeNested;
49   typedef SparseMatrixBase<SparseView > Base;
50 public:
51   EIGEN_SPARSE_PUBLIC_INTERFACE(SparseView)
52   typedef typename internal::remove_all<MatrixType>::type NestedExpression;
53 
54   explicit SparseView(const MatrixType& mat, const Scalar& reference = Scalar(0),
55                       const RealScalar &epsilon = NumTraits<Scalar>::dummy_precision())
56     : m_matrix(mat), m_reference(reference), m_epsilon(epsilon) {}
57 
58   inline Index rows() const { return m_matrix.rows(); }
59   inline Index cols() const { return m_matrix.cols(); }
60 
61   inline Index innerSize() const { return m_matrix.innerSize(); }
62   inline Index outerSize() const { return m_matrix.outerSize(); }
63 
64   /** \returns the nested expression */
65   const typename internal::remove_all<MatrixTypeNested>::type&
66   nestedExpression() const { return m_matrix; }
67 
68   Scalar reference() const { return m_reference; }
69   RealScalar epsilon() const { return m_epsilon; }
70 
71 protected:
72   MatrixTypeNested m_matrix;
73   Scalar m_reference;
74   RealScalar m_epsilon;
75 };
76 
77 namespace internal {
78 
79 // TODO find a way to unify the two following variants
80 // This is tricky because implementing an inner iterator on top of an IndexBased evaluator is
81 // not easy because the evaluators do not expose the sizes of the underlying expression.
82 
83 template<typename ArgType>
84 struct unary_evaluator<SparseView<ArgType>, IteratorBased>
85   : public evaluator_base<SparseView<ArgType> >
86 {
87     typedef typename evaluator<ArgType>::InnerIterator EvalIterator;
88   public:
89     typedef SparseView<ArgType> XprType;
90 
91     class InnerIterator : public EvalIterator
92     {
93         typedef typename XprType::Scalar Scalar;
94       public:
95 
96         EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& sve, Index outer)
97           : EvalIterator(sve.m_argImpl,outer), m_view(sve.m_view)
98         {
99           incrementToNonZero();
100         }
101 
102         EIGEN_STRONG_INLINE InnerIterator& operator++()
103         {
104           EvalIterator::operator++();
105           incrementToNonZero();
106           return *this;
107         }
108 
109         using EvalIterator::value;
110 
111       protected:
112         const XprType &m_view;
113 
114       private:
115         void incrementToNonZero()
116         {
117           while((bool(*this)) && internal::isMuchSmallerThan(value(), m_view.reference(), m_view.epsilon()))
118           {
119             EvalIterator::operator++();
120           }
121         }
122     };
123 
124     enum {
125       CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
126       Flags = XprType::Flags
127     };
128 
129     explicit unary_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_view(xpr) {}
130 
131   protected:
132     evaluator<ArgType> m_argImpl;
133     const XprType &m_view;
134 };
135 
136 template<typename ArgType>
137 struct unary_evaluator<SparseView<ArgType>, IndexBased>
138   : public evaluator_base<SparseView<ArgType> >
139 {
140   public:
141     typedef SparseView<ArgType> XprType;
142   protected:
143     enum { IsRowMajor = (XprType::Flags&RowMajorBit)==RowMajorBit };
144     typedef typename XprType::Scalar Scalar;
145     typedef typename XprType::StorageIndex StorageIndex;
146   public:
147 
148     class InnerIterator
149     {
150       public:
151 
152         EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& sve, Index outer)
153           : m_sve(sve), m_inner(0), m_outer(outer), m_end(sve.m_view.innerSize())
154         {
155           incrementToNonZero();
156         }
157 
158         EIGEN_STRONG_INLINE InnerIterator& operator++()
159         {
160           m_inner++;
161           incrementToNonZero();
162           return *this;
163         }
164 
165         EIGEN_STRONG_INLINE Scalar value() const
166         {
167           return (IsRowMajor) ? m_sve.m_argImpl.coeff(m_outer, m_inner)
168                               : m_sve.m_argImpl.coeff(m_inner, m_outer);
169         }
170 
171         EIGEN_STRONG_INLINE StorageIndex index() const { return m_inner; }
172         inline Index row() const { return IsRowMajor ? m_outer : index(); }
173         inline Index col() const { return IsRowMajor ? index() : m_outer; }
174 
175         EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; }
176 
177       protected:
178         const unary_evaluator &m_sve;
179         Index m_inner;
180         const Index m_outer;
181         const Index m_end;
182 
183       private:
184         void incrementToNonZero()
185         {
186           while((bool(*this)) && internal::isMuchSmallerThan(value(), m_sve.m_view.reference(), m_sve.m_view.epsilon()))
187           {
188             m_inner++;
189           }
190         }
191     };
192 
193     enum {
194       CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
195       Flags = XprType::Flags
196     };
197 
198     explicit unary_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_view(xpr) {}
199 
200   protected:
201     evaluator<ArgType> m_argImpl;
202     const XprType &m_view;
203 };
204 
205 } // end namespace internal
206 
207 /** \ingroup SparseCore_Module
208   *
209   * \returns a sparse expression of the dense expression \c *this with values smaller than
210   * \a reference * \a epsilon removed.
211   *
212   * This method is typically used when prototyping to convert a quickly assembled dense Matrix \c D to a SparseMatrix \c S:
213   * \code
214   * MatrixXd D(n,m);
215   * SparseMatrix<double> S;
216   * S = D.sparseView();             // suppress numerical zeros (exact)
217   * S = D.sparseView(reference);
218   * S = D.sparseView(reference,epsilon);
219   * \endcode
220   * where \a reference is a meaningful non zero reference value,
221   * and \a epsilon is a tolerance factor defaulting to NumTraits<Scalar>::dummy_precision().
222   *
223   * \sa SparseMatrixBase::pruned(), class SparseView */
224 template<typename Derived>
225 const SparseView<Derived> MatrixBase<Derived>::sparseView(const Scalar& reference,
226                                                           const typename NumTraits<Scalar>::Real& epsilon) const
227 {
228   return SparseView<Derived>(derived(), reference, epsilon);
229 }
230 
231 /** \returns an expression of \c *this with values smaller than
232   * \a reference * \a epsilon removed.
233   *
234   * This method is typically used in conjunction with the product of two sparse matrices
235   * to automatically prune the smallest values as follows:
236   * \code
237   * C = (A*B).pruned();             // suppress numerical zeros (exact)
238   * C = (A*B).pruned(ref);
239   * C = (A*B).pruned(ref,epsilon);
240   * \endcode
241   * where \c ref is a meaningful non zero reference value.
242   * */
243 template<typename Derived>
244 const SparseView<Derived>
245 SparseMatrixBase<Derived>::pruned(const Scalar& reference,
246                                   const RealScalar& epsilon) const
247 {
248   return SparseView<Derived>(derived(), reference, epsilon);
249 }
250 
251 } // end namespace Eigen
252 
253 #endif
254