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 <[email protected]>
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() = typename MatrixType::Scalar(0);
35 }
36 symmat = C + A;
37 }
38
39 }
40
41 /** \ingroup OrderingMethods_Module
42 * \class AMDOrdering
43 *
44 * Functor computing the \em approximate \em minimum \em degree ordering
45 * If the matrix is not structurally symmetric, an ordering of A^T+A is computed
46 * \tparam StorageIndex The type of indices of the matrix
47 * \sa COLAMDOrdering
48 */
49 template <typename StorageIndex>
50 class AMDOrdering
51 {
52 public:
53 typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
54
55 /** Compute the permutation vector from a sparse matrix
56 * This routine is much faster if the input matrix is column-major
57 */
58 template <typename MatrixType>
operator()59 void operator()(const MatrixType& mat, PermutationType& perm)
60 {
61 // Compute the symmetric pattern
62 SparseMatrix<typename MatrixType::Scalar, ColMajor, StorageIndex> symm;
63 internal::ordering_helper_at_plus_a(mat,symm);
64
65 // Call the AMD routine
66 //m_mat.prune(keep_diag());
67 internal::minimum_degree_ordering(symm, perm);
68 }
69
70 /** Compute the permutation with a selfadjoint matrix */
71 template <typename SrcType, unsigned int SrcUpLo>
operator()72 void operator()(const SparseSelfAdjointView<SrcType, SrcUpLo>& mat, PermutationType& perm)
73 {
74 SparseMatrix<typename SrcType::Scalar, ColMajor, StorageIndex> C; C = mat;
75
76 // Call the AMD routine
77 // m_mat.prune(keep_diag()); //Remove the diagonal elements
78 internal::minimum_degree_ordering(C, perm);
79 }
80 };
81
82 /** \ingroup OrderingMethods_Module
83 * \class NaturalOrdering
84 *
85 * Functor computing the natural ordering (identity)
86 *
87 * \note Returns an empty permutation matrix
88 * \tparam StorageIndex The type of indices of the matrix
89 */
90 template <typename StorageIndex>
91 class NaturalOrdering
92 {
93 public:
94 typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
95
96 /** Compute the permutation vector from a column-major sparse matrix */
97 template <typename MatrixType>
operator()98 void operator()(const MatrixType& /*mat*/, PermutationType& perm)
99 {
100 perm.resize(0);
101 }
102
103 };
104
105 /** \ingroup OrderingMethods_Module
106 * \class COLAMDOrdering
107 *
108 * \tparam StorageIndex The type of indices of the matrix
109 *
110 * Functor computing the \em column \em approximate \em minimum \em degree ordering
111 * The matrix should be in column-major and \b compressed format (see SparseMatrix::makeCompressed()).
112 */
113 template<typename StorageIndex>
114 class COLAMDOrdering
115 {
116 public:
117 typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
118 typedef Matrix<StorageIndex, Dynamic, 1> IndexVector;
119
120 /** Compute the permutation vector \a perm form the sparse matrix \a mat
121 * \warning The input sparse matrix \a mat must be in compressed mode (see SparseMatrix::makeCompressed()).
122 */
123 template <typename MatrixType>
operator()124 void operator() (const MatrixType& mat, PermutationType& perm)
125 {
126 eigen_assert(mat.isCompressed() && "COLAMDOrdering requires a sparse matrix in compressed mode. Call .makeCompressed() before passing it to COLAMDOrdering");
127
128 StorageIndex m = StorageIndex(mat.rows());
129 StorageIndex n = StorageIndex(mat.cols());
130 StorageIndex nnz = StorageIndex(mat.nonZeros());
131 // Get the recommended value of Alen to be used by colamd
132 StorageIndex Alen = internal::Colamd::recommended(nnz, m, n);
133 // Set the default parameters
134 double knobs [internal::Colamd::NKnobs];
135 StorageIndex stats [internal::Colamd::NStats];
136 internal::Colamd::set_defaults(knobs);
137
138 IndexVector p(n+1), A(Alen);
139 for(StorageIndex i=0; i <= n; i++) p(i) = mat.outerIndexPtr()[i];
140 for(StorageIndex i=0; i < nnz; i++) A(i) = mat.innerIndexPtr()[i];
141 // Call Colamd routine to compute the ordering
142 StorageIndex info = internal::Colamd::compute_ordering(m, n, Alen, A.data(), p.data(), knobs, stats);
143 EIGEN_UNUSED_VARIABLE(info);
144 eigen_assert( info && "COLAMD failed " );
145
146 perm.resize(n);
147 for (StorageIndex i = 0; i < n; i++) perm.indices()(p(i)) = i;
148 }
149 };
150
151 } // end namespace Eigen
152
153 #endif
154