• 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) 2009-2014 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_SPARSE_SELFADJOINTVIEW_H
11 #define EIGEN_SPARSE_SELFADJOINTVIEW_H
12 
13 namespace Eigen {
14 
15 /** \ingroup SparseCore_Module
16   * \class SparseSelfAdjointView
17   *
18   * \brief Pseudo expression to manipulate a triangular sparse matrix as a selfadjoint matrix.
19   *
20   * \param MatrixType the type of the dense matrix storing the coefficients
21   * \param Mode can be either \c #Lower or \c #Upper
22   *
23   * This class is an expression of a sefladjoint matrix from a triangular part of a matrix
24   * with given dense storage of the coefficients. It is the return type of MatrixBase::selfadjointView()
25   * and most of the time this is the only way that it is used.
26   *
27   * \sa SparseMatrixBase::selfadjointView()
28   */
29 namespace internal {
30 
31 template<typename MatrixType, unsigned int Mode>
32 struct traits<SparseSelfAdjointView<MatrixType,Mode> > : traits<MatrixType> {
33 };
34 
35 template<int SrcMode,int DstMode,typename MatrixType,int DestOrder>
36 void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm = 0);
37 
38 template<int Mode,typename MatrixType,int DestOrder>
39 void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm = 0);
40 
41 }
42 
43 template<typename MatrixType, unsigned int _Mode> class SparseSelfAdjointView
44   : public EigenBase<SparseSelfAdjointView<MatrixType,_Mode> >
45 {
46   public:
47 
48     enum {
49       Mode = _Mode,
50       TransposeMode = ((Mode & Upper) ? Lower : 0) | ((Mode & Lower) ? Upper : 0),
51       RowsAtCompileTime = internal::traits<SparseSelfAdjointView>::RowsAtCompileTime,
52       ColsAtCompileTime = internal::traits<SparseSelfAdjointView>::ColsAtCompileTime
53     };
54 
55     typedef EigenBase<SparseSelfAdjointView> Base;
56     typedef typename MatrixType::Scalar Scalar;
57     typedef typename MatrixType::StorageIndex StorageIndex;
58     typedef Matrix<StorageIndex,Dynamic,1> VectorI;
59     typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
60     typedef typename internal::remove_all<MatrixTypeNested>::type _MatrixTypeNested;
61 
62     explicit inline SparseSelfAdjointView(MatrixType& matrix) : m_matrix(matrix)
63     {
64       eigen_assert(rows()==cols() && "SelfAdjointView is only for squared matrices");
65     }
66 
67     inline Index rows() const { return m_matrix.rows(); }
68     inline Index cols() const { return m_matrix.cols(); }
69 
70     /** \internal \returns a reference to the nested matrix */
71     const _MatrixTypeNested& matrix() const { return m_matrix; }
72     typename internal::remove_reference<MatrixTypeNested>::type& matrix() { return m_matrix; }
73 
74     /** \returns an expression of the matrix product between a sparse self-adjoint matrix \c *this and a sparse matrix \a rhs.
75       *
76       * Note that there is no algorithmic advantage of performing such a product compared to a general sparse-sparse matrix product.
77       * Indeed, the SparseSelfadjointView operand is first copied into a temporary SparseMatrix before computing the product.
78       */
79     template<typename OtherDerived>
80     Product<SparseSelfAdjointView, OtherDerived>
81     operator*(const SparseMatrixBase<OtherDerived>& rhs) const
82     {
83       return Product<SparseSelfAdjointView, OtherDerived>(*this, rhs.derived());
84     }
85 
86     /** \returns an expression of the matrix product between a sparse matrix \a lhs and a sparse self-adjoint matrix \a rhs.
87       *
88       * Note that there is no algorithmic advantage of performing such a product compared to a general sparse-sparse matrix product.
89       * Indeed, the SparseSelfadjointView operand is first copied into a temporary SparseMatrix before computing the product.
90       */
91     template<typename OtherDerived> friend
92     Product<OtherDerived, SparseSelfAdjointView>
93     operator*(const SparseMatrixBase<OtherDerived>& lhs, const SparseSelfAdjointView& rhs)
94     {
95       return Product<OtherDerived, SparseSelfAdjointView>(lhs.derived(), rhs);
96     }
97 
98     /** Efficient sparse self-adjoint matrix times dense vector/matrix product */
99     template<typename OtherDerived>
100     Product<SparseSelfAdjointView,OtherDerived>
101     operator*(const MatrixBase<OtherDerived>& rhs) const
102     {
103       return Product<SparseSelfAdjointView,OtherDerived>(*this, rhs.derived());
104     }
105 
106     /** Efficient dense vector/matrix times sparse self-adjoint matrix product */
107     template<typename OtherDerived> friend
108     Product<OtherDerived,SparseSelfAdjointView>
109     operator*(const MatrixBase<OtherDerived>& lhs, const SparseSelfAdjointView& rhs)
110     {
111       return Product<OtherDerived,SparseSelfAdjointView>(lhs.derived(), rhs);
112     }
113 
114     /** Perform a symmetric rank K update of the selfadjoint matrix \c *this:
115       * \f$ this = this + \alpha ( u u^* ) \f$ where \a u is a vector or matrix.
116       *
117       * \returns a reference to \c *this
118       *
119       * To perform \f$ this = this + \alpha ( u^* u ) \f$ you can simply
120       * call this function with u.adjoint().
121       */
122     template<typename DerivedU>
123     SparseSelfAdjointView& rankUpdate(const SparseMatrixBase<DerivedU>& u, const Scalar& alpha = Scalar(1));
124 
125     /** \returns an expression of P H P^-1 */
126     // TODO implement twists in a more evaluator friendly fashion
127     SparseSymmetricPermutationProduct<_MatrixTypeNested,Mode> twistedBy(const PermutationMatrix<Dynamic,Dynamic,StorageIndex>& perm) const
128     {
129       return SparseSymmetricPermutationProduct<_MatrixTypeNested,Mode>(m_matrix, perm);
130     }
131 
132     template<typename SrcMatrixType,int SrcMode>
133     SparseSelfAdjointView& operator=(const SparseSymmetricPermutationProduct<SrcMatrixType,SrcMode>& permutedMatrix)
134     {
135       internal::call_assignment_no_alias_no_transpose(*this, permutedMatrix);
136       return *this;
137     }
138 
139     SparseSelfAdjointView& operator=(const SparseSelfAdjointView& src)
140     {
141       PermutationMatrix<Dynamic,Dynamic,StorageIndex> pnull;
142       return *this = src.twistedBy(pnull);
143     }
144 
145     template<typename SrcMatrixType,unsigned int SrcMode>
146     SparseSelfAdjointView& operator=(const SparseSelfAdjointView<SrcMatrixType,SrcMode>& src)
147     {
148       PermutationMatrix<Dynamic,Dynamic,StorageIndex> pnull;
149       return *this = src.twistedBy(pnull);
150     }
151 
152     void resize(Index rows, Index cols)
153     {
154       EIGEN_ONLY_USED_FOR_DEBUG(rows);
155       EIGEN_ONLY_USED_FOR_DEBUG(cols);
156       eigen_assert(rows == this->rows() && cols == this->cols()
157                 && "SparseSelfadjointView::resize() does not actually allow to resize.");
158     }
159 
160   protected:
161 
162     MatrixTypeNested m_matrix;
163     //mutable VectorI m_countPerRow;
164     //mutable VectorI m_countPerCol;
165   private:
166     template<typename Dest> void evalTo(Dest &) const;
167 };
168 
169 /***************************************************************************
170 * Implementation of SparseMatrixBase methods
171 ***************************************************************************/
172 
173 template<typename Derived>
174 template<unsigned int UpLo>
175 typename SparseMatrixBase<Derived>::template ConstSelfAdjointViewReturnType<UpLo>::Type SparseMatrixBase<Derived>::selfadjointView() const
176 {
177   return SparseSelfAdjointView<const Derived, UpLo>(derived());
178 }
179 
180 template<typename Derived>
181 template<unsigned int UpLo>
182 typename SparseMatrixBase<Derived>::template SelfAdjointViewReturnType<UpLo>::Type SparseMatrixBase<Derived>::selfadjointView()
183 {
184   return SparseSelfAdjointView<Derived, UpLo>(derived());
185 }
186 
187 /***************************************************************************
188 * Implementation of SparseSelfAdjointView methods
189 ***************************************************************************/
190 
191 template<typename MatrixType, unsigned int Mode>
192 template<typename DerivedU>
193 SparseSelfAdjointView<MatrixType,Mode>&
194 SparseSelfAdjointView<MatrixType,Mode>::rankUpdate(const SparseMatrixBase<DerivedU>& u, const Scalar& alpha)
195 {
196   SparseMatrix<Scalar,(MatrixType::Flags&RowMajorBit)?RowMajor:ColMajor> tmp = u * u.adjoint();
197   if(alpha==Scalar(0))
198     m_matrix = tmp.template triangularView<Mode>();
199   else
200     m_matrix += alpha * tmp.template triangularView<Mode>();
201 
202   return *this;
203 }
204 
205 namespace internal {
206 
207 // TODO currently a selfadjoint expression has the form SelfAdjointView<.,.>
208 //      in the future selfadjoint-ness should be defined by the expression traits
209 //      such that Transpose<SelfAdjointView<.,.> > is valid. (currently TriangularBase::transpose() is overloaded to make it work)
210 template<typename MatrixType, unsigned int Mode>
211 struct evaluator_traits<SparseSelfAdjointView<MatrixType,Mode> >
212 {
213   typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
214   typedef SparseSelfAdjointShape Shape;
215 };
216 
217 struct SparseSelfAdjoint2Sparse {};
218 
219 template<> struct AssignmentKind<SparseShape,SparseSelfAdjointShape> { typedef SparseSelfAdjoint2Sparse Kind; };
220 template<> struct AssignmentKind<SparseSelfAdjointShape,SparseShape> { typedef Sparse2Sparse Kind; };
221 
222 template< typename DstXprType, typename SrcXprType, typename Functor>
223 struct Assignment<DstXprType, SrcXprType, Functor, SparseSelfAdjoint2Sparse>
224 {
225   typedef typename DstXprType::StorageIndex StorageIndex;
226   typedef internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> AssignOpType;
227 
228   template<typename DestScalar,int StorageOrder>
229   static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src, const AssignOpType&/*func*/)
230   {
231     internal::permute_symm_to_fullsymm<SrcXprType::Mode>(src.matrix(), dst);
232   }
233 
234   // FIXME: the handling of += and -= in sparse matrices should be cleanup so that next two overloads could be reduced to:
235   template<typename DestScalar,int StorageOrder,typename AssignFunc>
236   static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src, const AssignFunc& func)
237   {
238     SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
239     run(tmp, src, AssignOpType());
240     call_assignment_no_alias_no_transpose(dst, tmp, func);
241   }
242 
243   template<typename DestScalar,int StorageOrder>
244   static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src,
245                   const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>& /* func */)
246   {
247     SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
248     run(tmp, src, AssignOpType());
249     dst += tmp;
250   }
251 
252   template<typename DestScalar,int StorageOrder>
253   static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src,
254                   const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>& /* func */)
255   {
256     SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
257     run(tmp, src, AssignOpType());
258     dst -= tmp;
259   }
260 
261   template<typename DestScalar>
262   static void run(DynamicSparseMatrix<DestScalar,ColMajor,StorageIndex>& dst, const SrcXprType &src, const AssignOpType&/*func*/)
263   {
264     // TODO directly evaluate into dst;
265     SparseMatrix<DestScalar,ColMajor,StorageIndex> tmp(dst.rows(),dst.cols());
266     internal::permute_symm_to_fullsymm<SrcXprType::Mode>(src.matrix(), tmp);
267     dst = tmp;
268   }
269 };
270 
271 } // end namespace internal
272 
273 /***************************************************************************
274 * Implementation of sparse self-adjoint time dense matrix
275 ***************************************************************************/
276 
277 namespace internal {
278 
279 template<int Mode, typename SparseLhsType, typename DenseRhsType, typename DenseResType, typename AlphaType>
280 inline void sparse_selfadjoint_time_dense_product(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const AlphaType& alpha)
281 {
282   EIGEN_ONLY_USED_FOR_DEBUG(alpha);
283 
284   typedef typename internal::nested_eval<SparseLhsType,DenseRhsType::MaxColsAtCompileTime>::type SparseLhsTypeNested;
285   typedef typename internal::remove_all<SparseLhsTypeNested>::type SparseLhsTypeNestedCleaned;
286   typedef evaluator<SparseLhsTypeNestedCleaned> LhsEval;
287   typedef typename LhsEval::InnerIterator LhsIterator;
288   typedef typename SparseLhsType::Scalar LhsScalar;
289 
290   enum {
291     LhsIsRowMajor = (LhsEval::Flags&RowMajorBit)==RowMajorBit,
292     ProcessFirstHalf =
293               ((Mode&(Upper|Lower))==(Upper|Lower))
294           || ( (Mode&Upper) && !LhsIsRowMajor)
295           || ( (Mode&Lower) && LhsIsRowMajor),
296     ProcessSecondHalf = !ProcessFirstHalf
297   };
298 
299   SparseLhsTypeNested lhs_nested(lhs);
300   LhsEval lhsEval(lhs_nested);
301 
302   // work on one column at once
303   for (Index k=0; k<rhs.cols(); ++k)
304   {
305     for (Index j=0; j<lhs.outerSize(); ++j)
306     {
307       LhsIterator i(lhsEval,j);
308       // handle diagonal coeff
309       if (ProcessSecondHalf)
310       {
311         while (i && i.index()<j) ++i;
312         if(i && i.index()==j)
313         {
314           res(j,k) += alpha * i.value() * rhs(j,k);
315           ++i;
316         }
317       }
318 
319       // premultiplied rhs for scatters
320       typename ScalarBinaryOpTraits<AlphaType, typename DenseRhsType::Scalar>::ReturnType rhs_j(alpha*rhs(j,k));
321       // accumulator for partial scalar product
322       typename DenseResType::Scalar res_j(0);
323       for(; (ProcessFirstHalf ? i && i.index() < j : i) ; ++i)
324       {
325         LhsScalar lhs_ij = i.value();
326         if(!LhsIsRowMajor) lhs_ij = numext::conj(lhs_ij);
327         res_j += lhs_ij * rhs(i.index(),k);
328         res(i.index(),k) += numext::conj(lhs_ij) * rhs_j;
329       }
330       res(j,k) += alpha * res_j;
331 
332       // handle diagonal coeff
333       if (ProcessFirstHalf && i && (i.index()==j))
334         res(j,k) += alpha * i.value() * rhs(j,k);
335     }
336   }
337 }
338 
339 
340 template<typename LhsView, typename Rhs, int ProductType>
341 struct generic_product_impl<LhsView, Rhs, SparseSelfAdjointShape, DenseShape, ProductType>
342 : generic_product_impl_base<LhsView, Rhs, generic_product_impl<LhsView, Rhs, SparseSelfAdjointShape, DenseShape, ProductType> >
343 {
344   template<typename Dest>
345   static void scaleAndAddTo(Dest& dst, const LhsView& lhsView, const Rhs& rhs, const typename Dest::Scalar& alpha)
346   {
347     typedef typename LhsView::_MatrixTypeNested Lhs;
348     typedef typename nested_eval<Lhs,Dynamic>::type LhsNested;
349     typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
350     LhsNested lhsNested(lhsView.matrix());
351     RhsNested rhsNested(rhs);
352 
353     internal::sparse_selfadjoint_time_dense_product<LhsView::Mode>(lhsNested, rhsNested, dst, alpha);
354   }
355 };
356 
357 template<typename Lhs, typename RhsView, int ProductType>
358 struct generic_product_impl<Lhs, RhsView, DenseShape, SparseSelfAdjointShape, ProductType>
359 : generic_product_impl_base<Lhs, RhsView, generic_product_impl<Lhs, RhsView, DenseShape, SparseSelfAdjointShape, ProductType> >
360 {
361   template<typename Dest>
362   static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const RhsView& rhsView, const typename Dest::Scalar& alpha)
363   {
364     typedef typename RhsView::_MatrixTypeNested Rhs;
365     typedef typename nested_eval<Lhs,Dynamic>::type LhsNested;
366     typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
367     LhsNested lhsNested(lhs);
368     RhsNested rhsNested(rhsView.matrix());
369 
370     // transpose everything
371     Transpose<Dest> dstT(dst);
372     internal::sparse_selfadjoint_time_dense_product<RhsView::TransposeMode>(rhsNested.transpose(), lhsNested.transpose(), dstT, alpha);
373   }
374 };
375 
376 // NOTE: these two overloads are needed to evaluate the sparse selfadjoint view into a full sparse matrix
377 // TODO: maybe the copy could be handled by generic_product_impl so that these overloads would not be needed anymore
378 
379 template<typename LhsView, typename Rhs, int ProductTag>
380 struct product_evaluator<Product<LhsView, Rhs, DefaultProduct>, ProductTag, SparseSelfAdjointShape, SparseShape>
381   : public evaluator<typename Product<typename Rhs::PlainObject, Rhs, DefaultProduct>::PlainObject>
382 {
383   typedef Product<LhsView, Rhs, DefaultProduct> XprType;
384   typedef typename XprType::PlainObject PlainObject;
385   typedef evaluator<PlainObject> Base;
386 
387   product_evaluator(const XprType& xpr)
388     : m_lhs(xpr.lhs()), m_result(xpr.rows(), xpr.cols())
389   {
390     ::new (static_cast<Base*>(this)) Base(m_result);
391     generic_product_impl<typename Rhs::PlainObject, Rhs, SparseShape, SparseShape, ProductTag>::evalTo(m_result, m_lhs, xpr.rhs());
392   }
393 
394 protected:
395   typename Rhs::PlainObject m_lhs;
396   PlainObject m_result;
397 };
398 
399 template<typename Lhs, typename RhsView, int ProductTag>
400 struct product_evaluator<Product<Lhs, RhsView, DefaultProduct>, ProductTag, SparseShape, SparseSelfAdjointShape>
401   : public evaluator<typename Product<Lhs, typename Lhs::PlainObject, DefaultProduct>::PlainObject>
402 {
403   typedef Product<Lhs, RhsView, DefaultProduct> XprType;
404   typedef typename XprType::PlainObject PlainObject;
405   typedef evaluator<PlainObject> Base;
406 
407   product_evaluator(const XprType& xpr)
408     : m_rhs(xpr.rhs()), m_result(xpr.rows(), xpr.cols())
409   {
410     ::new (static_cast<Base*>(this)) Base(m_result);
411     generic_product_impl<Lhs, typename Lhs::PlainObject, SparseShape, SparseShape, ProductTag>::evalTo(m_result, xpr.lhs(), m_rhs);
412   }
413 
414 protected:
415   typename Lhs::PlainObject m_rhs;
416   PlainObject m_result;
417 };
418 
419 } // namespace internal
420 
421 /***************************************************************************
422 * Implementation of symmetric copies and permutations
423 ***************************************************************************/
424 namespace internal {
425 
426 template<int Mode,typename MatrixType,int DestOrder>
427 void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm)
428 {
429   typedef typename MatrixType::StorageIndex StorageIndex;
430   typedef typename MatrixType::Scalar Scalar;
431   typedef SparseMatrix<Scalar,DestOrder,StorageIndex> Dest;
432   typedef Matrix<StorageIndex,Dynamic,1> VectorI;
433   typedef evaluator<MatrixType> MatEval;
434   typedef typename evaluator<MatrixType>::InnerIterator MatIterator;
435 
436   MatEval matEval(mat);
437   Dest& dest(_dest.derived());
438   enum {
439     StorageOrderMatch = int(Dest::IsRowMajor) == int(MatrixType::IsRowMajor)
440   };
441 
442   Index size = mat.rows();
443   VectorI count;
444   count.resize(size);
445   count.setZero();
446   dest.resize(size,size);
447   for(Index j = 0; j<size; ++j)
448   {
449     Index jp = perm ? perm[j] : j;
450     for(MatIterator it(matEval,j); it; ++it)
451     {
452       Index i = it.index();
453       Index r = it.row();
454       Index c = it.col();
455       Index ip = perm ? perm[i] : i;
456       if(Mode==(Upper|Lower))
457         count[StorageOrderMatch ? jp : ip]++;
458       else if(r==c)
459         count[ip]++;
460       else if(( Mode==Lower && r>c) || ( Mode==Upper && r<c))
461       {
462         count[ip]++;
463         count[jp]++;
464       }
465     }
466   }
467   Index nnz = count.sum();
468 
469   // reserve space
470   dest.resizeNonZeros(nnz);
471   dest.outerIndexPtr()[0] = 0;
472   for(Index j=0; j<size; ++j)
473     dest.outerIndexPtr()[j+1] = dest.outerIndexPtr()[j] + count[j];
474   for(Index j=0; j<size; ++j)
475     count[j] = dest.outerIndexPtr()[j];
476 
477   // copy data
478   for(StorageIndex j = 0; j<size; ++j)
479   {
480     for(MatIterator it(matEval,j); it; ++it)
481     {
482       StorageIndex i = internal::convert_index<StorageIndex>(it.index());
483       Index r = it.row();
484       Index c = it.col();
485 
486       StorageIndex jp = perm ? perm[j] : j;
487       StorageIndex ip = perm ? perm[i] : i;
488 
489       if(Mode==(Upper|Lower))
490       {
491         Index k = count[StorageOrderMatch ? jp : ip]++;
492         dest.innerIndexPtr()[k] = StorageOrderMatch ? ip : jp;
493         dest.valuePtr()[k] = it.value();
494       }
495       else if(r==c)
496       {
497         Index k = count[ip]++;
498         dest.innerIndexPtr()[k] = ip;
499         dest.valuePtr()[k] = it.value();
500       }
501       else if(( (Mode&Lower)==Lower && r>c) || ( (Mode&Upper)==Upper && r<c))
502       {
503         if(!StorageOrderMatch)
504           std::swap(ip,jp);
505         Index k = count[jp]++;
506         dest.innerIndexPtr()[k] = ip;
507         dest.valuePtr()[k] = it.value();
508         k = count[ip]++;
509         dest.innerIndexPtr()[k] = jp;
510         dest.valuePtr()[k] = numext::conj(it.value());
511       }
512     }
513   }
514 }
515 
516 template<int _SrcMode,int _DstMode,typename MatrixType,int DstOrder>
517 void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DstOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm)
518 {
519   typedef typename MatrixType::StorageIndex StorageIndex;
520   typedef typename MatrixType::Scalar Scalar;
521   SparseMatrix<Scalar,DstOrder,StorageIndex>& dest(_dest.derived());
522   typedef Matrix<StorageIndex,Dynamic,1> VectorI;
523   typedef evaluator<MatrixType> MatEval;
524   typedef typename evaluator<MatrixType>::InnerIterator MatIterator;
525 
526   enum {
527     SrcOrder = MatrixType::IsRowMajor ? RowMajor : ColMajor,
528     StorageOrderMatch = int(SrcOrder) == int(DstOrder),
529     DstMode = DstOrder==RowMajor ? (_DstMode==Upper ? Lower : Upper) : _DstMode,
530     SrcMode = SrcOrder==RowMajor ? (_SrcMode==Upper ? Lower : Upper) : _SrcMode
531   };
532 
533   MatEval matEval(mat);
534 
535   Index size = mat.rows();
536   VectorI count(size);
537   count.setZero();
538   dest.resize(size,size);
539   for(StorageIndex j = 0; j<size; ++j)
540   {
541     StorageIndex jp = perm ? perm[j] : j;
542     for(MatIterator it(matEval,j); it; ++it)
543     {
544       StorageIndex i = it.index();
545       if((int(SrcMode)==int(Lower) && i<j) || (int(SrcMode)==int(Upper) && i>j))
546         continue;
547 
548       StorageIndex ip = perm ? perm[i] : i;
549       count[int(DstMode)==int(Lower) ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
550     }
551   }
552   dest.outerIndexPtr()[0] = 0;
553   for(Index j=0; j<size; ++j)
554     dest.outerIndexPtr()[j+1] = dest.outerIndexPtr()[j] + count[j];
555   dest.resizeNonZeros(dest.outerIndexPtr()[size]);
556   for(Index j=0; j<size; ++j)
557     count[j] = dest.outerIndexPtr()[j];
558 
559   for(StorageIndex j = 0; j<size; ++j)
560   {
561 
562     for(MatIterator it(matEval,j); it; ++it)
563     {
564       StorageIndex i = it.index();
565       if((int(SrcMode)==int(Lower) && i<j) || (int(SrcMode)==int(Upper) && i>j))
566         continue;
567 
568       StorageIndex jp = perm ? perm[j] : j;
569       StorageIndex ip = perm? perm[i] : i;
570 
571       Index k = count[int(DstMode)==int(Lower) ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
572       dest.innerIndexPtr()[k] = int(DstMode)==int(Lower) ? (std::max)(ip,jp) : (std::min)(ip,jp);
573 
574       if(!StorageOrderMatch) std::swap(ip,jp);
575       if( ((int(DstMode)==int(Lower) && ip<jp) || (int(DstMode)==int(Upper) && ip>jp)))
576         dest.valuePtr()[k] = numext::conj(it.value());
577       else
578         dest.valuePtr()[k] = it.value();
579     }
580   }
581 }
582 
583 }
584 
585 // TODO implement twists in a more evaluator friendly fashion
586 
587 namespace internal {
588 
589 template<typename MatrixType, int Mode>
590 struct traits<SparseSymmetricPermutationProduct<MatrixType,Mode> > : traits<MatrixType> {
591 };
592 
593 }
594 
595 template<typename MatrixType,int Mode>
596 class SparseSymmetricPermutationProduct
597   : public EigenBase<SparseSymmetricPermutationProduct<MatrixType,Mode> >
598 {
599   public:
600     typedef typename MatrixType::Scalar Scalar;
601     typedef typename MatrixType::StorageIndex StorageIndex;
602     enum {
603       RowsAtCompileTime = internal::traits<SparseSymmetricPermutationProduct>::RowsAtCompileTime,
604       ColsAtCompileTime = internal::traits<SparseSymmetricPermutationProduct>::ColsAtCompileTime
605     };
606   protected:
607     typedef PermutationMatrix<Dynamic,Dynamic,StorageIndex> Perm;
608   public:
609     typedef Matrix<StorageIndex,Dynamic,1> VectorI;
610     typedef typename MatrixType::Nested MatrixTypeNested;
611     typedef typename internal::remove_all<MatrixTypeNested>::type NestedExpression;
612 
613     SparseSymmetricPermutationProduct(const MatrixType& mat, const Perm& perm)
614       : m_matrix(mat), m_perm(perm)
615     {}
616 
617     inline Index rows() const { return m_matrix.rows(); }
618     inline Index cols() const { return m_matrix.cols(); }
619 
620     const NestedExpression& matrix() const { return m_matrix; }
621     const Perm& perm() const { return m_perm; }
622 
623   protected:
624     MatrixTypeNested m_matrix;
625     const Perm& m_perm;
626 
627 };
628 
629 namespace internal {
630 
631 template<typename DstXprType, typename MatrixType, int Mode, typename Scalar>
632 struct Assignment<DstXprType, SparseSymmetricPermutationProduct<MatrixType,Mode>, internal::assign_op<Scalar,typename MatrixType::Scalar>, Sparse2Sparse>
633 {
634   typedef SparseSymmetricPermutationProduct<MatrixType,Mode> SrcXprType;
635   typedef typename DstXprType::StorageIndex DstIndex;
636   template<int Options>
637   static void run(SparseMatrix<Scalar,Options,DstIndex> &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename MatrixType::Scalar> &)
638   {
639     // internal::permute_symm_to_fullsymm<Mode>(m_matrix,_dest,m_perm.indices().data());
640     SparseMatrix<Scalar,(Options&RowMajor)==RowMajor ? ColMajor : RowMajor, DstIndex> tmp;
641     internal::permute_symm_to_fullsymm<Mode>(src.matrix(),tmp,src.perm().indices().data());
642     dst = tmp;
643   }
644 
645   template<typename DestType,unsigned int DestMode>
646   static void run(SparseSelfAdjointView<DestType,DestMode>& dst, const SrcXprType &src, const internal::assign_op<Scalar,typename MatrixType::Scalar> &)
647   {
648     internal::permute_symm_to_symm<Mode,DestMode>(src.matrix(),dst.matrix(),src.perm().indices().data());
649   }
650 };
651 
652 } // end namespace internal
653 
654 } // end namespace Eigen
655 
656 #endif // EIGEN_SPARSE_SELFADJOINTVIEW_H
657