IncompleteLUT.h
Go to the documentation of this file.
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
5 // Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
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_INCOMPLETE_LUT_H
12 #define EIGEN_INCOMPLETE_LUT_H
13 
14 
15 namespace Eigen {
16 
17 namespace internal {
18 
28 template <typename VectorV, typename VectorI>
29 Index QuickSplit(VectorV &row, VectorI &ind, Index ncut)
30 {
31  typedef typename VectorV::RealScalar RealScalar;
32  using std::swap;
33  using std::abs;
34  Index mid;
35  Index n = row.size(); /* length of the vector */
36  Index first, last ;
37 
38  ncut--; /* to fit the zero-based indices */
39  first = 0;
40  last = n-1;
41  if (ncut < first || ncut > last ) return 0;
42 
43  do {
44  mid = first;
45  RealScalar abskey = abs(row(mid));
46  for (Index j = first + 1; j <= last; j++) {
47  if ( abs(row(j)) > abskey) {
48  ++mid;
49  swap(row(mid), row(j));
50  swap(ind(mid), ind(j));
51  }
52  }
53  /* Interchange for the pivot element */
54  swap(row(mid), row(first));
55  swap(ind(mid), ind(first));
56 
57  if (mid > ncut) last = mid - 1;
58  else if (mid < ncut ) first = mid + 1;
59  } while (mid != ncut );
60 
61  return 0; /* mid is equal to ncut */
62 }
63 
64 }// end namespace internal
65 
98 template <typename _Scalar, typename _StorageIndex = int>
99 class IncompleteLUT : public SparseSolverBase<IncompleteLUT<_Scalar, _StorageIndex> >
100 {
101  protected:
103  using Base::m_isInitialized;
104  public:
105  typedef _Scalar Scalar;
106  typedef _StorageIndex StorageIndex;
111 
112  enum {
113  ColsAtCompileTime = Dynamic,
114  MaxColsAtCompileTime = Dynamic
115  };
116 
117  public:
118 
120  : m_droptol(NumTraits<Scalar>::dummy_precision()), m_fillfactor(10),
121  m_analysisIsOk(false), m_factorizationIsOk(false)
122  {}
123 
124  template<typename MatrixType>
125  explicit IncompleteLUT(const MatrixType& mat, const RealScalar& droptol=NumTraits<Scalar>::dummy_precision(), int fillfactor = 10)
126  : m_droptol(droptol),m_fillfactor(fillfactor),
127  m_analysisIsOk(false),m_factorizationIsOk(false)
128  {
129  eigen_assert(fillfactor != 0);
130  compute(mat);
131  }
132 
133  Index rows() const { return m_lu.rows(); }
134 
135  Index cols() const { return m_lu.cols(); }
136 
143  {
144  eigen_assert(m_isInitialized && "IncompleteLUT is not initialized.");
145  return m_info;
146  }
147 
148  template<typename MatrixType>
149  void analyzePattern(const MatrixType& amat);
150 
151  template<typename MatrixType>
152  void factorize(const MatrixType& amat);
153 
159  template<typename MatrixType>
160  IncompleteLUT& compute(const MatrixType& amat)
161  {
162  analyzePattern(amat);
163  factorize(amat);
164  return *this;
165  }
166 
167  void setDroptol(const RealScalar& droptol);
168  void setFillfactor(int fillfactor);
169 
170  template<typename Rhs, typename Dest>
171  void _solve_impl(const Rhs& b, Dest& x) const
172  {
173  x = m_Pinv * b;
174  x = m_lu.template triangularView<UnitLower>().solve(x);
175  x = m_lu.template triangularView<Upper>().solve(x);
176  x = m_P * x;
177  }
178 
179 protected:
180 
182  struct keep_diag {
183  inline bool operator() (const Index& row, const Index& col, const Scalar&) const
184  {
185  return row!=col;
186  }
187  };
188 
189 protected:
190 
191  FactorType m_lu;
192  RealScalar m_droptol;
199 };
200 
205 template<typename Scalar, typename StorageIndex>
206 void IncompleteLUT<Scalar,StorageIndex>::setDroptol(const RealScalar& droptol)
207 {
208  this->m_droptol = droptol;
209 }
210 
215 template<typename Scalar, typename StorageIndex>
217 {
218  this->m_fillfactor = fillfactor;
219 }
220 
221 template <typename Scalar, typename StorageIndex>
222 template<typename _MatrixType>
224 {
225  // Compute the Fill-reducing permutation
226  // Since ILUT does not perform any numerical pivoting,
227  // it is highly preferable to keep the diagonal through symmetric permutations.
228 #ifndef EIGEN_MPL2_ONLY
229  // To this end, let's symmetrize the pattern and perform AMD on it.
232  // FIXME for a matrix with nearly symmetric pattern, mat2+mat1 is the appropriate choice.
233  // on the other hand for a really non-symmetric pattern, mat2*mat1 should be prefered...
235  AMDOrdering<StorageIndex> ordering;
236  ordering(AtA,m_P);
237  m_Pinv = m_P.inverse(); // cache the inverse permutation
238 #else
239  // If AMD is not available, (MPL2-only), then let's use the slower COLAMD routine.
242  ordering(mat1,m_Pinv);
243  m_P = m_Pinv.inverse();
244 #endif
245 
246  m_analysisIsOk = true;
247  m_factorizationIsOk = false;
248  m_isInitialized = true;
249 }
250 
251 template <typename Scalar, typename StorageIndex>
252 template<typename _MatrixType>
254 {
255  using std::sqrt;
256  using std::swap;
257  using std::abs;
259 
260  eigen_assert((amat.rows() == amat.cols()) && "The factorization should be done on a square matrix");
261  Index n = amat.cols(); // Size of the matrix
262  m_lu.resize(n,n);
263  // Declare Working vectors and variables
264  Vector u(n) ; // real values of the row -- maximum size is n --
265  VectorI ju(n); // column position of the values in u -- maximum size is n
266  VectorI jr(n); // Indicate the position of the nonzero elements in the vector u -- A zero location is indicated by -1
267 
268  // Apply the fill-reducing permutation
269  eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
271  mat = amat.twistedBy(m_Pinv);
272 
273  // Initialization
274  jr.fill(-1);
275  ju.fill(0);
276  u.fill(0);
277 
278  // number of largest elements to keep in each row:
279  Index fill_in = (amat.nonZeros()*m_fillfactor)/n + 1;
280  if (fill_in > n) fill_in = n;
281 
282  // number of largest nonzero elements to keep in the L and the U part of the current row:
283  Index nnzL = fill_in/2;
284  Index nnzU = nnzL;
285  m_lu.reserve(n * (nnzL + nnzU + 1));
286 
287  // global loop over the rows of the sparse matrix
288  for (Index ii = 0; ii < n; ii++)
289  {
290  // 1 - copy the lower and the upper part of the row i of mat in the working vector u
291 
292  Index sizeu = 1; // number of nonzero elements in the upper part of the current row
293  Index sizel = 0; // number of nonzero elements in the lower part of the current row
294  ju(ii) = convert_index<StorageIndex>(ii);
295  u(ii) = 0;
296  jr(ii) = convert_index<StorageIndex>(ii);
297  RealScalar rownorm = 0;
298 
299  typename FactorType::InnerIterator j_it(mat, ii); // Iterate through the current row ii
300  for (; j_it; ++j_it)
301  {
302  Index k = j_it.index();
303  if (k < ii)
304  {
305  // copy the lower part
306  ju(sizel) = convert_index<StorageIndex>(k);
307  u(sizel) = j_it.value();
308  jr(k) = convert_index<StorageIndex>(sizel);
309  ++sizel;
310  }
311  else if (k == ii)
312  {
313  u(ii) = j_it.value();
314  }
315  else
316  {
317  // copy the upper part
318  Index jpos = ii + sizeu;
319  ju(jpos) = convert_index<StorageIndex>(k);
320  u(jpos) = j_it.value();
321  jr(k) = convert_index<StorageIndex>(jpos);
322  ++sizeu;
323  }
324  rownorm += numext::abs2(j_it.value());
325  }
326 
327  // 2 - detect possible zero row
328  if(rownorm==0)
329  {
330  m_info = NumericalIssue;
331  return;
332  }
333  // Take the 2-norm of the current row as a relative tolerance
334  rownorm = sqrt(rownorm);
335 
336  // 3 - eliminate the previous nonzero rows
337  Index jj = 0;
338  Index len = 0;
339  while (jj < sizel)
340  {
341  // In order to eliminate in the correct order,
342  // we must select first the smallest column index among ju(jj:sizel)
343  Index k;
344  Index minrow = ju.segment(jj,sizel-jj).minCoeff(&k); // k is relative to the segment
345  k += jj;
346  if (minrow != ju(jj))
347  {
348  // swap the two locations
349  Index j = ju(jj);
350  swap(ju(jj), ju(k));
351  jr(minrow) = convert_index<StorageIndex>(jj);
352  jr(j) = convert_index<StorageIndex>(k);
353  swap(u(jj), u(k));
354  }
355  // Reset this location
356  jr(minrow) = -1;
357 
358  // Start elimination
359  typename FactorType::InnerIterator ki_it(m_lu, minrow);
360  while (ki_it && ki_it.index() < minrow) ++ki_it;
361  eigen_internal_assert(ki_it && ki_it.col()==minrow);
362  Scalar fact = u(jj) / ki_it.value();
363 
364  // drop too small elements
365  if(abs(fact) <= m_droptol)
366  {
367  jj++;
368  continue;
369  }
370 
371  // linear combination of the current row ii and the row minrow
372  ++ki_it;
373  for (; ki_it; ++ki_it)
374  {
375  Scalar prod = fact * ki_it.value();
376  Index j = ki_it.index();
377  Index jpos = jr(j);
378  if (jpos == -1) // fill-in element
379  {
380  Index newpos;
381  if (j >= ii) // dealing with the upper part
382  {
383  newpos = ii + sizeu;
384  sizeu++;
385  eigen_internal_assert(sizeu<=n);
386  }
387  else // dealing with the lower part
388  {
389  newpos = sizel;
390  sizel++;
391  eigen_internal_assert(sizel<=ii);
392  }
393  ju(newpos) = convert_index<StorageIndex>(j);
394  u(newpos) = -prod;
395  jr(j) = convert_index<StorageIndex>(newpos);
396  }
397  else
398  u(jpos) -= prod;
399  }
400  // store the pivot element
401  u(len) = fact;
402  ju(len) = convert_index<StorageIndex>(minrow);
403  ++len;
404 
405  jj++;
406  } // end of the elimination on the row ii
407 
408  // reset the upper part of the pointer jr to zero
409  for(Index k = 0; k <sizeu; k++) jr(ju(ii+k)) = -1;
410 
411  // 4 - partially sort and insert the elements in the m_lu matrix
412 
413  // sort the L-part of the row
414  sizel = len;
415  len = (std::min)(sizel, nnzL);
416  typename Vector::SegmentReturnType ul(u.segment(0, sizel));
417  typename VectorI::SegmentReturnType jul(ju.segment(0, sizel));
418  internal::QuickSplit(ul, jul, len);
419 
420  // store the largest m_fill elements of the L part
421  m_lu.startVec(ii);
422  for(Index k = 0; k < len; k++)
423  m_lu.insertBackByOuterInnerUnordered(ii,ju(k)) = u(k);
424 
425  // store the diagonal element
426  // apply a shifting rule to avoid zero pivots (we are doing an incomplete factorization)
427  if (u(ii) == Scalar(0))
428  u(ii) = sqrt(m_droptol) * rownorm;
429  m_lu.insertBackByOuterInnerUnordered(ii, ii) = u(ii);
430 
431  // sort the U-part of the row
432  // apply the dropping rule first
433  len = 0;
434  for(Index k = 1; k < sizeu; k++)
435  {
436  if(abs(u(ii+k)) > m_droptol * rownorm )
437  {
438  ++len;
439  u(ii + len) = u(ii + k);
440  ju(ii + len) = ju(ii + k);
441  }
442  }
443  sizeu = len + 1; // +1 to take into account the diagonal element
444  len = (std::min)(sizeu, nnzU);
445  typename Vector::SegmentReturnType uu(u.segment(ii+1, sizeu-1));
446  typename VectorI::SegmentReturnType juu(ju.segment(ii+1, sizeu-1));
447  internal::QuickSplit(uu, juu, len);
448 
449  // store the largest elements of the U part
450  for(Index k = ii + 1; k < ii + len; k++)
451  m_lu.insertBackByOuterInnerUnordered(ii,ju(k)) = u(k);
452  }
453  m_lu.finalize();
454  m_lu.makeCompressed();
455 
456  m_factorizationIsOk = true;
457  m_info = Success;
458 }
459 
460 } // end namespace Eigen
461 
462 #endif // EIGEN_INCOMPLETE_LUT_H
void setDroptol(const RealScalar &droptol)
IncompleteLUT(const MatrixType &mat, const RealScalar &droptol=NumTraits< Scalar >::dummy_precision(), int fillfactor=10)
VectorBlock< Derived > SegmentReturnType
Definition: BlockMethods.h:38
void factorize(const MatrixType &amat)
ComputationInfo m_info
SparseSolverBase< IncompleteLUT > Base
A base class for sparse solvers.
ComputationInfo info() const
Reports whether previous computation was successful.
EIGEN_DEVICE_FUNC const SqrtReturnType sqrt() const
Definition: LDLT.h:16
Holds information about the various numeric (i.e. scalar) types allowed by Eigen. ...
Definition: NumTraits.h:150
EIGEN_DEVICE_FUNC IndexDest convert_index(const IndexSrc &idx)
Definition: XprHelper.h:31
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const AbsReturnType abs() const
PermutationMatrix< Dynamic, Dynamic, StorageIndex > m_Pinv
Matrix< StorageIndex, Dynamic, 1 > VectorI
Index cols() const
void _solve_impl(const Rhs &b, Dest &x) const
Index rows() const
EIGEN_DEVICE_FUNC ColXpr col(Index i)
This is the const version of col().
Definition: BlockMethods.h:838
PermutationMatrix< Dynamic, Dynamic, StorageIndex > m_P
Index QuickSplit(VectorV &row, VectorI &ind, Index ncut)
Definition: IncompleteLUT.h:29
TransposeReturnType transpose()
EIGEN_DEFAULT_DENSE_INDEX_TYPE Index
The Index type as used for the API.
Definition: Meta.h:33
#define eigen_assert(x)
Definition: Macros.h:577
EIGEN_DEVICE_FUNC RowXpr row(Index i)
This is the const version of row(). */.
Definition: BlockMethods.h:859
Incomplete LU factorization with dual-threshold strategy.
Definition: IncompleteLUT.h:99
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Abs2ReturnType abs2() const
SparseSymmetricPermutationProduct< Derived, Upper|Lower > twistedBy(const PermutationMatrix< Dynamic, Dynamic, StorageIndex > &perm) const
Matrix< Scalar, Dynamic, 1 > Vector
_StorageIndex StorageIndex
SparseMatrix< Scalar, RowMajor, StorageIndex > FactorType
void analyzePattern(const MatrixType &amat)
NumTraits< Scalar >::Real RealScalar
IncompleteLUT & compute(const MatrixType &amat)
const int Dynamic
Definition: Constants.h:21
#define eigen_internal_assert(x)
Definition: Macros.h:583
ComputationInfo
Definition: Constants.h:430
EIGEN_DEVICE_FUNC const Scalar & b
void swap(mpfr::mpreal &x, mpfr::mpreal &y)
Definition: mpreal.h:2986
void swap(scoped_array< T > &a, scoped_array< T > &b)
Definition: Memory.h:602
void setFillfactor(int fillfactor)


hebiros
Author(s): Xavier Artache , Matthew Tesch
autogenerated on Thu Sep 3 2020 04:08:15