Eigen  3.2.7
 All Classes Namespaces Functions Variables Typedefs Enumerations Enumerator Friends Groups Pages
UmfPackSupport.h
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2008-2011 Gael Guennebaud <[email protected]>
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_UMFPACKSUPPORT_H
11 #define EIGEN_UMFPACKSUPPORT_H
12 
13 namespace Eigen {
14 
15 /* TODO extract L, extract U, compute det, etc... */
16 
17 // generic double/complex<double> wrapper functions:
18 
19 inline void umfpack_free_numeric(void **Numeric, double)
20 { umfpack_di_free_numeric(Numeric); *Numeric = 0; }
21 
22 inline void umfpack_free_numeric(void **Numeric, std::complex<double>)
23 { umfpack_zi_free_numeric(Numeric); *Numeric = 0; }
24 
25 inline void umfpack_free_symbolic(void **Symbolic, double)
26 { umfpack_di_free_symbolic(Symbolic); *Symbolic = 0; }
27 
28 inline void umfpack_free_symbolic(void **Symbolic, std::complex<double>)
29 { umfpack_zi_free_symbolic(Symbolic); *Symbolic = 0; }
30 
31 inline int umfpack_symbolic(int n_row,int n_col,
32  const int Ap[], const int Ai[], const double Ax[], void **Symbolic,
33  const double Control [UMFPACK_CONTROL], double Info [UMFPACK_INFO])
34 {
35  return umfpack_di_symbolic(n_row,n_col,Ap,Ai,Ax,Symbolic,Control,Info);
36 }
37 
38 inline int umfpack_symbolic(int n_row,int n_col,
39  const int Ap[], const int Ai[], const std::complex<double> Ax[], void **Symbolic,
40  const double Control [UMFPACK_CONTROL], double Info [UMFPACK_INFO])
41 {
42  return umfpack_zi_symbolic(n_row,n_col,Ap,Ai,&numext::real_ref(Ax[0]),0,Symbolic,Control,Info);
43 }
44 
45 inline int umfpack_numeric( const int Ap[], const int Ai[], const double Ax[],
46  void *Symbolic, void **Numeric,
47  const double Control[UMFPACK_CONTROL],double Info [UMFPACK_INFO])
48 {
49  return umfpack_di_numeric(Ap,Ai,Ax,Symbolic,Numeric,Control,Info);
50 }
51 
52 inline int umfpack_numeric( const int Ap[], const int Ai[], const std::complex<double> Ax[],
53  void *Symbolic, void **Numeric,
54  const double Control[UMFPACK_CONTROL],double Info [UMFPACK_INFO])
55 {
56  return umfpack_zi_numeric(Ap,Ai,&numext::real_ref(Ax[0]),0,Symbolic,Numeric,Control,Info);
57 }
58 
59 inline int umfpack_solve( int sys, const int Ap[], const int Ai[], const double Ax[],
60  double X[], const double B[], void *Numeric,
61  const double Control[UMFPACK_CONTROL], double Info[UMFPACK_INFO])
62 {
63  return umfpack_di_solve(sys,Ap,Ai,Ax,X,B,Numeric,Control,Info);
64 }
65 
66 inline int umfpack_solve( int sys, const int Ap[], const int Ai[], const std::complex<double> Ax[],
67  std::complex<double> X[], const std::complex<double> B[], void *Numeric,
68  const double Control[UMFPACK_CONTROL], double Info[UMFPACK_INFO])
69 {
70  return umfpack_zi_solve(sys,Ap,Ai,&numext::real_ref(Ax[0]),0,&numext::real_ref(X[0]),0,&numext::real_ref(B[0]),0,Numeric,Control,Info);
71 }
72 
73 inline int umfpack_get_lunz(int *lnz, int *unz, int *n_row, int *n_col, int *nz_udiag, void *Numeric, double)
74 {
75  return umfpack_di_get_lunz(lnz,unz,n_row,n_col,nz_udiag,Numeric);
76 }
77 
78 inline int umfpack_get_lunz(int *lnz, int *unz, int *n_row, int *n_col, int *nz_udiag, void *Numeric, std::complex<double>)
79 {
80  return umfpack_zi_get_lunz(lnz,unz,n_row,n_col,nz_udiag,Numeric);
81 }
82 
83 inline int umfpack_get_numeric(int Lp[], int Lj[], double Lx[], int Up[], int Ui[], double Ux[],
84  int P[], int Q[], double Dx[], int *do_recip, double Rs[], void *Numeric)
85 {
86  return umfpack_di_get_numeric(Lp,Lj,Lx,Up,Ui,Ux,P,Q,Dx,do_recip,Rs,Numeric);
87 }
88 
89 inline int umfpack_get_numeric(int Lp[], int Lj[], std::complex<double> Lx[], int Up[], int Ui[], std::complex<double> Ux[],
90  int P[], int Q[], std::complex<double> Dx[], int *do_recip, double Rs[], void *Numeric)
91 {
92  double& lx0_real = numext::real_ref(Lx[0]);
93  double& ux0_real = numext::real_ref(Ux[0]);
94  double& dx0_real = numext::real_ref(Dx[0]);
95  return umfpack_zi_get_numeric(Lp,Lj,Lx?&lx0_real:0,0,Up,Ui,Ux?&ux0_real:0,0,P,Q,
96  Dx?&dx0_real:0,0,do_recip,Rs,Numeric);
97 }
98 
99 inline int umfpack_get_determinant(double *Mx, double *Ex, void *NumericHandle, double User_Info [UMFPACK_INFO])
100 {
101  return umfpack_di_get_determinant(Mx,Ex,NumericHandle,User_Info);
102 }
103 
104 inline int umfpack_get_determinant(std::complex<double> *Mx, double *Ex, void *NumericHandle, double User_Info [UMFPACK_INFO])
105 {
106  double& mx_real = numext::real_ref(*Mx);
107  return umfpack_zi_get_determinant(&mx_real,0,Ex,NumericHandle,User_Info);
108 }
109 
110 namespace internal {
111  template<typename T> struct umfpack_helper_is_sparse_plain : false_type {};
112  template<typename Scalar, int Options, typename StorageIndex>
113  struct umfpack_helper_is_sparse_plain<SparseMatrix<Scalar,Options,StorageIndex> >
114  : true_type {};
115  template<typename Scalar, int Options, typename StorageIndex>
116  struct umfpack_helper_is_sparse_plain<MappedSparseMatrix<Scalar,Options,StorageIndex> >
117  : true_type {};
118 }
119 
133 template<typename _MatrixType>
134 class UmfPackLU : internal::noncopyable
135 {
136  public:
137  typedef _MatrixType MatrixType;
138  typedef typename MatrixType::Scalar Scalar;
139  typedef typename MatrixType::RealScalar RealScalar;
140  typedef typename MatrixType::Index Index;
146 
147  public:
148 
149  UmfPackLU() { init(); }
150 
151  UmfPackLU(const MatrixType& matrix)
152  {
153  init();
154  compute(matrix);
155  }
156 
157  ~UmfPackLU()
158  {
159  if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
160  if(m_numeric) umfpack_free_numeric(&m_numeric,Scalar());
161  }
162 
163  inline Index rows() const { return m_copyMatrix.rows(); }
164  inline Index cols() const { return m_copyMatrix.cols(); }
165 
172  {
173  eigen_assert(m_isInitialized && "Decomposition is not initialized.");
174  return m_info;
175  }
176 
177  inline const LUMatrixType& matrixL() const
178  {
179  if (m_extractedDataAreDirty) extractData();
180  return m_l;
181  }
182 
183  inline const LUMatrixType& matrixU() const
184  {
185  if (m_extractedDataAreDirty) extractData();
186  return m_u;
187  }
188 
189  inline const IntColVectorType& permutationP() const
190  {
191  if (m_extractedDataAreDirty) extractData();
192  return m_p;
193  }
194 
195  inline const IntRowVectorType& permutationQ() const
196  {
197  if (m_extractedDataAreDirty) extractData();
198  return m_q;
199  }
200 
205  template<typename InputMatrixType>
206  void compute(const InputMatrixType& matrix)
207  {
208  if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
209  if(m_numeric) umfpack_free_numeric(&m_numeric,Scalar());
210  grapInput(matrix.derived());
211  analyzePattern_impl();
212  factorize_impl();
213  }
214 
219  template<typename Rhs>
220  inline const internal::solve_retval<UmfPackLU, Rhs> solve(const MatrixBase<Rhs>& b) const
221  {
222  eigen_assert(m_isInitialized && "UmfPackLU is not initialized.");
223  eigen_assert(rows()==b.rows()
224  && "UmfPackLU::solve(): invalid number of rows of the right hand side matrix b");
225  return internal::solve_retval<UmfPackLU, Rhs>(*this, b.derived());
226  }
227 
232  template<typename Rhs>
233  inline const internal::sparse_solve_retval<UmfPackLU, Rhs> solve(const SparseMatrixBase<Rhs>& b) const
234  {
235  eigen_assert(m_isInitialized && "UmfPackLU is not initialized.");
236  eigen_assert(rows()==b.rows()
237  && "UmfPackLU::solve(): invalid number of rows of the right hand side matrix b");
238  return internal::sparse_solve_retval<UmfPackLU, Rhs>(*this, b.derived());
239  }
240 
247  template<typename InputMatrixType>
248  void analyzePattern(const InputMatrixType& matrix)
249  {
250  if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
251  if(m_numeric) umfpack_free_numeric(&m_numeric,Scalar());
252 
253  grapInput(matrix.derived());
254 
255  analyzePattern_impl();
256  }
257 
264  template<typename InputMatrixType>
265  void factorize(const InputMatrixType& matrix)
266  {
267  eigen_assert(m_analysisIsOk && "UmfPackLU: you must first call analyzePattern()");
268  if(m_numeric)
269  umfpack_free_numeric(&m_numeric,Scalar());
270 
271  grapInput(matrix.derived());
272 
273  factorize_impl();
274  }
275 
276  #ifndef EIGEN_PARSED_BY_DOXYGEN
277 
278  template<typename BDerived,typename XDerived>
279  bool _solve(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const;
280  #endif
281 
282  Scalar determinant() const;
283 
284  void extractData() const;
285 
286  protected:
287 
288  void init()
289  {
290  m_info = InvalidInput;
291  m_isInitialized = false;
292  m_numeric = 0;
293  m_symbolic = 0;
294  m_outerIndexPtr = 0;
295  m_innerIndexPtr = 0;
296  m_valuePtr = 0;
297  m_extractedDataAreDirty = true;
298  }
299 
300  template<typename InputMatrixType>
301  void grapInput_impl(const InputMatrixType& mat, internal::true_type)
302  {
303  m_copyMatrix.resize(mat.rows(), mat.cols());
304  if( ((MatrixType::Flags&RowMajorBit)==RowMajorBit) || sizeof(typename MatrixType::Index)!=sizeof(int) || !mat.isCompressed() )
305  {
306  // non supported input -> copy
307  m_copyMatrix = mat;
308  m_outerIndexPtr = m_copyMatrix.outerIndexPtr();
309  m_innerIndexPtr = m_copyMatrix.innerIndexPtr();
310  m_valuePtr = m_copyMatrix.valuePtr();
311  }
312  else
313  {
314  m_outerIndexPtr = mat.outerIndexPtr();
315  m_innerIndexPtr = mat.innerIndexPtr();
316  m_valuePtr = mat.valuePtr();
317  }
318  }
319 
320  template<typename InputMatrixType>
321  void grapInput_impl(const InputMatrixType& mat, internal::false_type)
322  {
323  m_copyMatrix = mat;
324  m_outerIndexPtr = m_copyMatrix.outerIndexPtr();
325  m_innerIndexPtr = m_copyMatrix.innerIndexPtr();
326  m_valuePtr = m_copyMatrix.valuePtr();
327  }
328 
329  template<typename InputMatrixType>
330  void grapInput(const InputMatrixType& mat)
331  {
332  grapInput_impl(mat, internal::umfpack_helper_is_sparse_plain<InputMatrixType>());
333  }
334 
335  void analyzePattern_impl()
336  {
337  int errorCode = 0;
338  errorCode = umfpack_symbolic(m_copyMatrix.rows(), m_copyMatrix.cols(), m_outerIndexPtr, m_innerIndexPtr, m_valuePtr,
339  &m_symbolic, 0, 0);
340 
341  m_isInitialized = true;
342  m_info = errorCode ? InvalidInput : Success;
343  m_analysisIsOk = true;
344  m_factorizationIsOk = false;
345  m_extractedDataAreDirty = true;
346  }
347 
348  void factorize_impl()
349  {
350  int errorCode;
351  errorCode = umfpack_numeric(m_outerIndexPtr, m_innerIndexPtr, m_valuePtr,
352  m_symbolic, &m_numeric, 0, 0);
353 
354  m_info = errorCode ? NumericalIssue : Success;
355  m_factorizationIsOk = true;
356  m_extractedDataAreDirty = true;
357  }
358 
359  // cached data to reduce reallocation, etc.
360  mutable LUMatrixType m_l;
361  mutable LUMatrixType m_u;
362  mutable IntColVectorType m_p;
363  mutable IntRowVectorType m_q;
364 
365  UmfpackMatrixType m_copyMatrix;
366  const Scalar* m_valuePtr;
367  const int* m_outerIndexPtr;
368  const int* m_innerIndexPtr;
369  void* m_numeric;
370  void* m_symbolic;
371 
372  mutable ComputationInfo m_info;
373  bool m_isInitialized;
374  int m_factorizationIsOk;
375  int m_analysisIsOk;
376  mutable bool m_extractedDataAreDirty;
377 
378  private:
379  UmfPackLU(UmfPackLU& ) { }
380 };
381 
382 
383 template<typename MatrixType>
384 void UmfPackLU<MatrixType>::extractData() const
385 {
386  if (m_extractedDataAreDirty)
387  {
388  // get size of the data
389  int lnz, unz, rows, cols, nz_udiag;
390  umfpack_get_lunz(&lnz, &unz, &rows, &cols, &nz_udiag, m_numeric, Scalar());
391 
392  // allocate data
393  m_l.resize(rows,(std::min)(rows,cols));
394  m_l.resizeNonZeros(lnz);
395 
396  m_u.resize((std::min)(rows,cols),cols);
397  m_u.resizeNonZeros(unz);
398 
399  m_p.resize(rows);
400  m_q.resize(cols);
401 
402  // extract
403  umfpack_get_numeric(m_l.outerIndexPtr(), m_l.innerIndexPtr(), m_l.valuePtr(),
404  m_u.outerIndexPtr(), m_u.innerIndexPtr(), m_u.valuePtr(),
405  m_p.data(), m_q.data(), 0, 0, 0, m_numeric);
406 
407  m_extractedDataAreDirty = false;
408  }
409 }
410 
411 template<typename MatrixType>
412 typename UmfPackLU<MatrixType>::Scalar UmfPackLU<MatrixType>::determinant() const
413 {
414  Scalar det;
415  umfpack_get_determinant(&det, 0, m_numeric, 0);
416  return det;
417 }
418 
419 template<typename MatrixType>
420 template<typename BDerived,typename XDerived>
421 bool UmfPackLU<MatrixType>::_solve(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const
422 {
423  const int rhsCols = b.cols();
424  eigen_assert((BDerived::Flags&RowMajorBit)==0 && "UmfPackLU backend does not support non col-major rhs yet");
425  eigen_assert((XDerived::Flags&RowMajorBit)==0 && "UmfPackLU backend does not support non col-major result yet");
426  eigen_assert(b.derived().data() != x.derived().data() && " Umfpack does not support inplace solve");
427 
428  int errorCode;
429  for (int j=0; j<rhsCols; ++j)
430  {
431  errorCode = umfpack_solve(UMFPACK_A,
432  m_outerIndexPtr, m_innerIndexPtr, m_valuePtr,
433  &x.col(j).coeffRef(0), &b.const_cast_derived().col(j).coeffRef(0), m_numeric, 0, 0);
434  if (errorCode!=0)
435  return false;
436  }
437 
438  return true;
439 }
440 
441 
442 namespace internal {
443 
444 template<typename _MatrixType, typename Rhs>
445 struct solve_retval<UmfPackLU<_MatrixType>, Rhs>
446  : solve_retval_base<UmfPackLU<_MatrixType>, Rhs>
447 {
448  typedef UmfPackLU<_MatrixType> Dec;
449  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
450 
451  template<typename Dest> void evalTo(Dest& dst) const
452  {
453  dec()._solve(rhs(),dst);
454  }
455 };
456 
457 template<typename _MatrixType, typename Rhs>
458 struct sparse_solve_retval<UmfPackLU<_MatrixType>, Rhs>
459  : sparse_solve_retval_base<UmfPackLU<_MatrixType>, Rhs>
460 {
461  typedef UmfPackLU<_MatrixType> Dec;
462  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
463 
464  template<typename Dest> void evalTo(Dest& dst) const
465  {
466  this->defaultEvalTo(dst);
467  }
468 };
469 
470 } // end namespace internal
471 
472 } // end namespace Eigen
473 
474 #endif // EIGEN_UMFPACKSUPPORT_H
Index rows() const
Definition: SparseMatrix.h:119
A sparse LU factorization and solver based on UmfPack.
Definition: UmfPackSupport.h:134
Index cols() const
Definition: SparseMatrix.h:121
const Scalar * valuePtr() const
Definition: SparseMatrix.h:131
void resize(Index rows, Index cols)
Definition: SparseMatrix.h:596
void factorize(const InputMatrixType &matrix)
Definition: UmfPackSupport.h:265
Definition: Constants.h:378
const internal::sparse_solve_retval< UmfPackLU, Rhs > solve(const SparseMatrixBase< Rhs > &b) const
Definition: UmfPackSupport.h:233
const Index * outerIndexPtr() const
Definition: SparseMatrix.h:149
void compute(const InputMatrixType &matrix)
Definition: UmfPackSupport.h:206
Base class of any sparse matrices or sparse expressions.
Definition: ForwardDeclarations.h:239
Derived & derived()
Definition: EigenBase.h:34
Definition: Constants.h:383
ComputationInfo info() const
Reports whether previous computation was successful.
Definition: UmfPackSupport.h:171
Definition: Constants.h:376
const unsigned int RowMajorBit
Definition: Constants.h:53
void analyzePattern(const InputMatrixType &matrix)
Definition: UmfPackSupport.h:248
Index rows() const
Definition: SparseMatrixBase.h:159
ComputationInfo
Definition: Constants.h:374
Base class for all dense matrices, vectors, and expressions.
Definition: MatrixBase.h:48
const internal::solve_retval< UmfPackLU, Rhs > solve(const MatrixBase< Rhs > &b) const
Definition: UmfPackSupport.h:220
const Index * innerIndexPtr() const
Definition: SparseMatrix.h:140