1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
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_SPARSESPARSEPRODUCTWITHPRUNING_H
11 #define EIGEN_SPARSESPARSEPRODUCTWITHPRUNING_H
12
13 namespace Eigen {
14
15 namespace internal {
16
17
18 // perform a pseudo in-place sparse * sparse product assuming all matrices are col major
19 template<typename Lhs, typename Rhs, typename ResultType>
sparse_sparse_product_with_pruning_impl(const Lhs & lhs,const Rhs & rhs,ResultType & res,const typename ResultType::RealScalar & tolerance)20 static void sparse_sparse_product_with_pruning_impl(const Lhs& lhs, const Rhs& rhs, ResultType& res, const typename ResultType::RealScalar& tolerance)
21 {
22 // return sparse_sparse_product_with_pruning_impl2(lhs,rhs,res);
23
24 typedef typename remove_all<Rhs>::type::Scalar RhsScalar;
25 typedef typename remove_all<ResultType>::type::Scalar ResScalar;
26 typedef typename remove_all<Lhs>::type::StorageIndex StorageIndex;
27
28 // make sure to call innerSize/outerSize since we fake the storage order.
29 Index rows = lhs.innerSize();
30 Index cols = rhs.outerSize();
31 //Index size = lhs.outerSize();
32 eigen_assert(lhs.outerSize() == rhs.innerSize());
33
34 // allocate a temporary buffer
35 AmbiVector<ResScalar,StorageIndex> tempVector(rows);
36
37 // mimics a resizeByInnerOuter:
38 if(ResultType::IsRowMajor)
39 res.resize(cols, rows);
40 else
41 res.resize(rows, cols);
42
43 evaluator<Lhs> lhsEval(lhs);
44 evaluator<Rhs> rhsEval(rhs);
45
46 // estimate the number of non zero entries
47 // given a rhs column containing Y non zeros, we assume that the respective Y columns
48 // of the lhs differs in average of one non zeros, thus the number of non zeros for
49 // the product of a rhs column with the lhs is X+Y where X is the average number of non zero
50 // per column of the lhs.
51 // Therefore, we have nnz(lhs*rhs) = nnz(lhs) + nnz(rhs)
52 Index estimated_nnz_prod = lhsEval.nonZerosEstimate() + rhsEval.nonZerosEstimate();
53
54 res.reserve(estimated_nnz_prod);
55 double ratioColRes = double(estimated_nnz_prod)/(double(lhs.rows())*double(rhs.cols()));
56 for (Index j=0; j<cols; ++j)
57 {
58 // FIXME:
59 //double ratioColRes = (double(rhs.innerVector(j).nonZeros()) + double(lhs.nonZeros())/double(lhs.cols()))/double(lhs.rows());
60 // let's do a more accurate determination of the nnz ratio for the current column j of res
61 tempVector.init(ratioColRes);
62 tempVector.setZero();
63 for (typename evaluator<Rhs>::InnerIterator rhsIt(rhsEval, j); rhsIt; ++rhsIt)
64 {
65 // FIXME should be written like this: tmp += rhsIt.value() * lhs.col(rhsIt.index())
66 tempVector.restart();
67 RhsScalar x = rhsIt.value();
68 for (typename evaluator<Lhs>::InnerIterator lhsIt(lhsEval, rhsIt.index()); lhsIt; ++lhsIt)
69 {
70 tempVector.coeffRef(lhsIt.index()) += lhsIt.value() * x;
71 }
72 }
73 res.startVec(j);
74 for (typename AmbiVector<ResScalar,StorageIndex>::Iterator it(tempVector,tolerance); it; ++it)
75 res.insertBackByOuterInner(j,it.index()) = it.value();
76 }
77 res.finalize();
78 }
79
80 template<typename Lhs, typename Rhs, typename ResultType,
81 int LhsStorageOrder = traits<Lhs>::Flags&RowMajorBit,
82 int RhsStorageOrder = traits<Rhs>::Flags&RowMajorBit,
83 int ResStorageOrder = traits<ResultType>::Flags&RowMajorBit>
84 struct sparse_sparse_product_with_pruning_selector;
85
86 template<typename Lhs, typename Rhs, typename ResultType>
87 struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,ColMajor,ColMajor>
88 {
89 typedef typename ResultType::RealScalar RealScalar;
90
91 static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
92 {
93 typename remove_all<ResultType>::type _res(res.rows(), res.cols());
94 internal::sparse_sparse_product_with_pruning_impl<Lhs,Rhs,ResultType>(lhs, rhs, _res, tolerance);
95 res.swap(_res);
96 }
97 };
98
99 template<typename Lhs, typename Rhs, typename ResultType>
100 struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,ColMajor,RowMajor>
101 {
102 typedef typename ResultType::RealScalar RealScalar;
103 static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
104 {
105 // we need a col-major matrix to hold the result
106 typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> SparseTemporaryType;
107 SparseTemporaryType _res(res.rows(), res.cols());
108 internal::sparse_sparse_product_with_pruning_impl<Lhs,Rhs,SparseTemporaryType>(lhs, rhs, _res, tolerance);
109 res = _res;
110 }
111 };
112
113 template<typename Lhs, typename Rhs, typename ResultType>
114 struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,RowMajor,RowMajor,RowMajor>
115 {
116 typedef typename ResultType::RealScalar RealScalar;
117 static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
118 {
119 // let's transpose the product to get a column x column product
120 typename remove_all<ResultType>::type _res(res.rows(), res.cols());
121 internal::sparse_sparse_product_with_pruning_impl<Rhs,Lhs,ResultType>(rhs, lhs, _res, tolerance);
122 res.swap(_res);
123 }
124 };
125
126 template<typename Lhs, typename Rhs, typename ResultType>
127 struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,RowMajor,RowMajor,ColMajor>
128 {
129 typedef typename ResultType::RealScalar RealScalar;
130 static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
131 {
132 typedef SparseMatrix<typename Lhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixLhs;
133 typedef SparseMatrix<typename Rhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixRhs;
134 ColMajorMatrixLhs colLhs(lhs);
135 ColMajorMatrixRhs colRhs(rhs);
136 internal::sparse_sparse_product_with_pruning_impl<ColMajorMatrixLhs,ColMajorMatrixRhs,ResultType>(colLhs, colRhs, res, tolerance);
137
138 // let's transpose the product to get a column x column product
139 // typedef SparseMatrix<typename ResultType::Scalar> SparseTemporaryType;
140 // SparseTemporaryType _res(res.cols(), res.rows());
141 // sparse_sparse_product_with_pruning_impl<Rhs,Lhs,SparseTemporaryType>(rhs, lhs, _res);
142 // res = _res.transpose();
143 }
144 };
145
146 template<typename Lhs, typename Rhs, typename ResultType>
147 struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,RowMajor,RowMajor>
148 {
149 typedef typename ResultType::RealScalar RealScalar;
150 static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
151 {
152 typedef SparseMatrix<typename Lhs::Scalar,RowMajor,typename Lhs::StorageIndex> RowMajorMatrixLhs;
153 RowMajorMatrixLhs rowLhs(lhs);
154 sparse_sparse_product_with_pruning_selector<RowMajorMatrixLhs,Rhs,ResultType,RowMajor,RowMajor>(rowLhs,rhs,res,tolerance);
155 }
156 };
157
158 template<typename Lhs, typename Rhs, typename ResultType>
159 struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,RowMajor,ColMajor,RowMajor>
160 {
161 typedef typename ResultType::RealScalar RealScalar;
162 static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
163 {
164 typedef SparseMatrix<typename Rhs::Scalar,RowMajor,typename Lhs::StorageIndex> RowMajorMatrixRhs;
165 RowMajorMatrixRhs rowRhs(rhs);
166 sparse_sparse_product_with_pruning_selector<Lhs,RowMajorMatrixRhs,ResultType,RowMajor,RowMajor,RowMajor>(lhs,rowRhs,res,tolerance);
167 }
168 };
169
170 template<typename Lhs, typename Rhs, typename ResultType>
171 struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,RowMajor,ColMajor>
172 {
173 typedef typename ResultType::RealScalar RealScalar;
174 static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
175 {
176 typedef SparseMatrix<typename Rhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixRhs;
177 ColMajorMatrixRhs colRhs(rhs);
178 internal::sparse_sparse_product_with_pruning_impl<Lhs,ColMajorMatrixRhs,ResultType>(lhs, colRhs, res, tolerance);
179 }
180 };
181
182 template<typename Lhs, typename Rhs, typename ResultType>
183 struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,RowMajor,ColMajor,ColMajor>
184 {
185 typedef typename ResultType::RealScalar RealScalar;
186 static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
187 {
188 typedef SparseMatrix<typename Lhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixLhs;
189 ColMajorMatrixLhs colLhs(lhs);
190 internal::sparse_sparse_product_with_pruning_impl<ColMajorMatrixLhs,Rhs,ResultType>(colLhs, rhs, res, tolerance);
191 }
192 };
193
194 } // end namespace internal
195
196 } // end namespace Eigen
197
198 #endif // EIGEN_SPARSESPARSEPRODUCTWITHPRUNING_H
199