1// This file is part of Eigen, a lightweight C++ template library
2// for linear algebra.
3//
4// Copyright (C) 2008-2011 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_CONSERVATIVESPARSESPARSEPRODUCT_H
11#define EIGEN_CONSERVATIVESPARSESPARSEPRODUCT_H
12
13namespace Eigen {
14
15namespace internal {
16
17template<typename Lhs, typename Rhs, typename ResultType>
18static void conservative_sparse_sparse_product_impl(const Lhs& lhs, const Rhs& rhs, ResultType& res)
19{
20  typedef typename remove_all<Lhs>::type::Scalar Scalar;
21  typedef typename remove_all<Lhs>::type::Index Index;
22
23  // make sure to call innerSize/outerSize since we fake the storage order.
24  Index rows = lhs.innerSize();
25  Index cols = rhs.outerSize();
26  eigen_assert(lhs.outerSize() == rhs.innerSize());
27
28  std::vector<bool> mask(rows,false);
29  Matrix<Scalar,Dynamic,1> values(rows);
30  Matrix<Index,Dynamic,1>  indices(rows);
31
32  // estimate the number of non zero entries
33  // given a rhs column containing Y non zeros, we assume that the respective Y columns
34  // of the lhs differs in average of one non zeros, thus the number of non zeros for
35  // the product of a rhs column with the lhs is X+Y where X is the average number of non zero
36  // per column of the lhs.
37  // Therefore, we have nnz(lhs*rhs) = nnz(lhs) + nnz(rhs)
38  Index estimated_nnz_prod = lhs.nonZeros() + rhs.nonZeros();
39
40  res.setZero();
41  res.reserve(Index(estimated_nnz_prod));
42  // we compute each column of the result, one after the other
43  for (Index j=0; j<cols; ++j)
44  {
45
46    res.startVec(j);
47    Index nnz = 0;
48    for (typename Rhs::InnerIterator rhsIt(rhs, j); rhsIt; ++rhsIt)
49    {
50      Scalar y = rhsIt.value();
51      Index k = rhsIt.index();
52      for (typename Lhs::InnerIterator lhsIt(lhs, k); lhsIt; ++lhsIt)
53      {
54        Index i = lhsIt.index();
55        Scalar x = lhsIt.value();
56        if(!mask[i])
57        {
58          mask[i] = true;
59          values[i] = x * y;
60          indices[nnz] = i;
61          ++nnz;
62        }
63        else
64          values[i] += x * y;
65      }
66    }
67
68    // unordered insertion
69    for(Index k=0; k<nnz; ++k)
70    {
71      Index i = indices[k];
72      res.insertBackByOuterInnerUnordered(j,i) = values[i];
73      mask[i] = false;
74    }
75
76#if 0
77    // alternative ordered insertion code:
78
79    Index t200 = rows/(log2(200)*1.39);
80    Index t = (rows*100)/139;
81
82    // FIXME reserve nnz non zeros
83    // FIXME implement fast sort algorithms for very small nnz
84    // if the result is sparse enough => use a quick sort
85    // otherwise => loop through the entire vector
86    // In order to avoid to perform an expensive log2 when the
87    // result is clearly very sparse we use a linear bound up to 200.
88    //if((nnz<200 && nnz<t200) || nnz * log2(nnz) < t)
89    //res.startVec(j);
90    if(true)
91    {
92      if(nnz>1) std::sort(indices.data(),indices.data()+nnz);
93      for(Index k=0; k<nnz; ++k)
94      {
95        Index i = indices[k];
96        res.insertBackByOuterInner(j,i) = values[i];
97        mask[i] = false;
98      }
99    }
100    else
101    {
102      // dense path
103      for(Index i=0; i<rows; ++i)
104      {
105        if(mask[i])
106        {
107          mask[i] = false;
108          res.insertBackByOuterInner(j,i) = values[i];
109        }
110      }
111    }
112#endif
113
114  }
115  res.finalize();
116}
117
118
119} // end namespace internal
120
121namespace internal {
122
123template<typename Lhs, typename Rhs, typename ResultType,
124  int LhsStorageOrder = (traits<Lhs>::Flags&RowMajorBit) ? RowMajor : ColMajor,
125  int RhsStorageOrder = (traits<Rhs>::Flags&RowMajorBit) ? RowMajor : ColMajor,
126  int ResStorageOrder = (traits<ResultType>::Flags&RowMajorBit) ? RowMajor : ColMajor>
127struct conservative_sparse_sparse_product_selector;
128
129template<typename Lhs, typename Rhs, typename ResultType>
130struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,ColMajor,ColMajor,ColMajor>
131{
132  typedef typename remove_all<Lhs>::type LhsCleaned;
133  typedef typename LhsCleaned::Scalar Scalar;
134
135  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
136  {
137    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
138    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
139    ColMajorMatrix resCol(lhs.rows(),rhs.cols());
140    internal::conservative_sparse_sparse_product_impl<Lhs,Rhs,ColMajorMatrix>(lhs, rhs, resCol);
141    // sort the non zeros:
142    RowMajorMatrix resRow(resCol);
143    res = resRow;
144  }
145};
146
147template<typename Lhs, typename Rhs, typename ResultType>
148struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,ColMajor,ColMajor>
149{
150  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
151  {
152     typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
153     RowMajorMatrix rhsRow = rhs;
154     RowMajorMatrix resRow(lhs.rows(), rhs.cols());
155     internal::conservative_sparse_sparse_product_impl<RowMajorMatrix,Lhs,RowMajorMatrix>(rhsRow, lhs, resRow);
156     res = resRow;
157  }
158};
159
160template<typename Lhs, typename Rhs, typename ResultType>
161struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,ColMajor,RowMajor,ColMajor>
162{
163  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
164  {
165    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
166    RowMajorMatrix lhsRow = lhs;
167    RowMajorMatrix resRow(lhs.rows(), rhs.cols());
168    internal::conservative_sparse_sparse_product_impl<Rhs,RowMajorMatrix,RowMajorMatrix>(rhs, lhsRow, resRow);
169    res = resRow;
170  }
171};
172
173template<typename Lhs, typename Rhs, typename ResultType>
174struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,RowMajor,ColMajor>
175{
176  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
177  {
178    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
179    RowMajorMatrix resRow(lhs.rows(), rhs.cols());
180    internal::conservative_sparse_sparse_product_impl<Rhs,Lhs,RowMajorMatrix>(rhs, lhs, resRow);
181    res = resRow;
182  }
183};
184
185
186template<typename Lhs, typename Rhs, typename ResultType>
187struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,ColMajor,ColMajor,RowMajor>
188{
189  typedef typename traits<typename remove_all<Lhs>::type>::Scalar Scalar;
190
191  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
192  {
193    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
194    ColMajorMatrix resCol(lhs.rows(), rhs.cols());
195    internal::conservative_sparse_sparse_product_impl<Lhs,Rhs,ColMajorMatrix>(lhs, rhs, resCol);
196    res = resCol;
197  }
198};
199
200template<typename Lhs, typename Rhs, typename ResultType>
201struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,ColMajor,RowMajor>
202{
203  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
204  {
205    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
206    ColMajorMatrix lhsCol = lhs;
207    ColMajorMatrix resCol(lhs.rows(), rhs.cols());
208    internal::conservative_sparse_sparse_product_impl<ColMajorMatrix,Rhs,ColMajorMatrix>(lhsCol, rhs, resCol);
209    res = resCol;
210  }
211};
212
213template<typename Lhs, typename Rhs, typename ResultType>
214struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,ColMajor,RowMajor,RowMajor>
215{
216  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
217  {
218    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
219    ColMajorMatrix rhsCol = rhs;
220    ColMajorMatrix resCol(lhs.rows(), rhs.cols());
221    internal::conservative_sparse_sparse_product_impl<Lhs,ColMajorMatrix,ColMajorMatrix>(lhs, rhsCol, resCol);
222    res = resCol;
223  }
224};
225
226template<typename Lhs, typename Rhs, typename ResultType>
227struct conservative_sparse_sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,RowMajor,RowMajor>
228{
229  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
230  {
231    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
232    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
233    RowMajorMatrix resRow(lhs.rows(),rhs.cols());
234    internal::conservative_sparse_sparse_product_impl<Rhs,Lhs,RowMajorMatrix>(rhs, lhs, resRow);
235    // sort the non zeros:
236    ColMajorMatrix resCol(resRow);
237    res = resCol;
238  }
239};
240
241} // end namespace internal
242
243} // end namespace Eigen
244
245#endif // EIGEN_CONSERVATIVESPARSESPARSEPRODUCT_H
246