• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 
2 // This file is part of Eigen, a lightweight C++ template library
3 // for linear algebra.
4 //
5 // Copyright (C) 2012  Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
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_ORDERING_H
12 #define EIGEN_ORDERING_H
13 
14 namespace Eigen {
15 
16 #include "Eigen_Colamd.h"
17 
18 namespace internal {
19 
20 /** \internal
21   * \ingroup OrderingMethods_Module
22   * \param[in] A the input non-symmetric matrix
23   * \param[out] symmat the symmetric pattern A^T+A from the input matrix \a A.
24   * FIXME: The values should not be considered here
25   */
26 template<typename MatrixType>
ordering_helper_at_plus_a(const MatrixType & A,MatrixType & symmat)27 void ordering_helper_at_plus_a(const MatrixType& A, MatrixType& symmat)
28 {
29   MatrixType C;
30   C = A.transpose(); // NOTE: Could be  costly
31   for (int i = 0; i < C.rows(); i++)
32   {
33       for (typename MatrixType::InnerIterator it(C, i); it; ++it)
34         it.valueRef() = 0.0;
35   }
36   symmat = C + A;
37 }
38 
39 }
40 
41 #ifndef EIGEN_MPL2_ONLY
42 
43 /** \ingroup OrderingMethods_Module
44   * \class AMDOrdering
45   *
46   * Functor computing the \em approximate \em minimum \em degree ordering
47   * If the matrix is not structurally symmetric, an ordering of A^T+A is computed
48   * \tparam  StorageIndex The type of indices of the matrix
49   * \sa COLAMDOrdering
50   */
51 template <typename StorageIndex>
52 class AMDOrdering
53 {
54   public:
55     typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
56 
57     /** Compute the permutation vector from a sparse matrix
58      * This routine is much faster if the input matrix is column-major
59      */
60     template <typename MatrixType>
operator()61     void operator()(const MatrixType& mat, PermutationType& perm)
62     {
63       // Compute the symmetric pattern
64       SparseMatrix<typename MatrixType::Scalar, ColMajor, StorageIndex> symm;
65       internal::ordering_helper_at_plus_a(mat,symm);
66 
67       // Call the AMD routine
68       //m_mat.prune(keep_diag());
69       internal::minimum_degree_ordering(symm, perm);
70     }
71 
72     /** Compute the permutation with a selfadjoint matrix */
73     template <typename SrcType, unsigned int SrcUpLo>
operator()74     void operator()(const SparseSelfAdjointView<SrcType, SrcUpLo>& mat, PermutationType& perm)
75     {
76       SparseMatrix<typename SrcType::Scalar, ColMajor, StorageIndex> C; C = mat;
77 
78       // Call the AMD routine
79       // m_mat.prune(keep_diag()); //Remove the diagonal elements
80       internal::minimum_degree_ordering(C, perm);
81     }
82 };
83 
84 #endif // EIGEN_MPL2_ONLY
85 
86 /** \ingroup OrderingMethods_Module
87   * \class NaturalOrdering
88   *
89   * Functor computing the natural ordering (identity)
90   *
91   * \note Returns an empty permutation matrix
92   * \tparam  StorageIndex The type of indices of the matrix
93   */
94 template <typename StorageIndex>
95 class NaturalOrdering
96 {
97   public:
98     typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
99 
100     /** Compute the permutation vector from a column-major sparse matrix */
101     template <typename MatrixType>
operator()102     void operator()(const MatrixType& /*mat*/, PermutationType& perm)
103     {
104       perm.resize(0);
105     }
106 
107 };
108 
109 /** \ingroup OrderingMethods_Module
110   * \class COLAMDOrdering
111   *
112   * \tparam  StorageIndex The type of indices of the matrix
113   *
114   * Functor computing the \em column \em approximate \em minimum \em degree ordering
115   * The matrix should be in column-major and \b compressed format (see SparseMatrix::makeCompressed()).
116   */
117 template<typename StorageIndex>
118 class COLAMDOrdering
119 {
120   public:
121     typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
122     typedef Matrix<StorageIndex, Dynamic, 1> IndexVector;
123 
124     /** Compute the permutation vector \a perm form the sparse matrix \a mat
125       * \warning The input sparse matrix \a mat must be in compressed mode (see SparseMatrix::makeCompressed()).
126       */
127     template <typename MatrixType>
operator()128     void operator() (const MatrixType& mat, PermutationType& perm)
129     {
130       eigen_assert(mat.isCompressed() && "COLAMDOrdering requires a sparse matrix in compressed mode. Call .makeCompressed() before passing it to COLAMDOrdering");
131 
132       StorageIndex m = StorageIndex(mat.rows());
133       StorageIndex n = StorageIndex(mat.cols());
134       StorageIndex nnz = StorageIndex(mat.nonZeros());
135       // Get the recommended value of Alen to be used by colamd
136       StorageIndex Alen = internal::colamd_recommended(nnz, m, n);
137       // Set the default parameters
138       double knobs [COLAMD_KNOBS];
139       StorageIndex stats [COLAMD_STATS];
140       internal::colamd_set_defaults(knobs);
141 
142       IndexVector p(n+1), A(Alen);
143       for(StorageIndex i=0; i <= n; i++)   p(i) = mat.outerIndexPtr()[i];
144       for(StorageIndex i=0; i < nnz; i++)  A(i) = mat.innerIndexPtr()[i];
145       // Call Colamd routine to compute the ordering
146       StorageIndex info = internal::colamd(m, n, Alen, A.data(), p.data(), knobs, stats);
147       EIGEN_UNUSED_VARIABLE(info);
148       eigen_assert( info && "COLAMD failed " );
149 
150       perm.resize(n);
151       for (StorageIndex i = 0; i < n; i++) perm.indices()(p(i)) = i;
152     }
153 };
154 
155 } // end namespace Eigen
156 
157 #endif
158