• 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) 2011 Gael Guennebaud <g.gael@free.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 #include "sparse.h"
11 #include <Eigen/SparseCore>
12 #include <sstream>
13 
14 template<typename Solver, typename Rhs, typename Guess,typename Result>
solve_with_guess(IterativeSolverBase<Solver> & solver,const MatrixBase<Rhs> & b,const Guess & g,Result & x)15 void solve_with_guess(IterativeSolverBase<Solver>& solver, const MatrixBase<Rhs>& b, const Guess& g, Result &x) {
16   if(internal::random<bool>())
17   {
18     // With a temporary through evaluator<SolveWithGuess>
19     x = solver.derived().solveWithGuess(b,g) + Result::Zero(x.rows(), x.cols());
20   }
21   else
22   {
23     // direct evaluation within x through Assignment<Result,SolveWithGuess>
24     x = solver.derived().solveWithGuess(b.derived(),g);
25   }
26 }
27 
28 template<typename Solver, typename Rhs, typename Guess,typename Result>
solve_with_guess(SparseSolverBase<Solver> & solver,const MatrixBase<Rhs> & b,const Guess &,Result & x)29 void solve_with_guess(SparseSolverBase<Solver>& solver, const MatrixBase<Rhs>& b, const Guess& , Result& x) {
30   if(internal::random<bool>())
31     x = solver.derived().solve(b) + Result::Zero(x.rows(), x.cols());
32   else
33     x = solver.derived().solve(b);
34 }
35 
36 template<typename Solver, typename Rhs, typename Guess,typename Result>
solve_with_guess(SparseSolverBase<Solver> & solver,const SparseMatrixBase<Rhs> & b,const Guess &,Result & x)37 void solve_with_guess(SparseSolverBase<Solver>& solver, const SparseMatrixBase<Rhs>& b, const Guess& , Result& x) {
38   x = solver.derived().solve(b);
39 }
40 
41 template<typename Solver, typename Rhs, typename DenseMat, typename DenseRhs>
check_sparse_solving(Solver & solver,const typename Solver::MatrixType & A,const Rhs & b,const DenseMat & dA,const DenseRhs & db)42 void check_sparse_solving(Solver& solver, const typename Solver::MatrixType& A, const Rhs& b, const DenseMat& dA, const DenseRhs& db)
43 {
44   typedef typename Solver::MatrixType Mat;
45   typedef typename Mat::Scalar Scalar;
46   typedef typename Mat::StorageIndex StorageIndex;
47 
48   DenseRhs refX = dA.householderQr().solve(db);
49   {
50     Rhs x(A.cols(), b.cols());
51     Rhs oldb = b;
52 
53     solver.compute(A);
54     if (solver.info() != Success)
55     {
56       std::cerr << "ERROR | sparse solver testing, factorization failed (" << typeid(Solver).name() << ")\n";
57       VERIFY(solver.info() == Success);
58     }
59     x = solver.solve(b);
60     if (solver.info() != Success)
61     {
62       std::cerr << "WARNING | sparse solver testing: solving failed (" << typeid(Solver).name() << ")\n";
63       return;
64     }
65     VERIFY(oldb.isApprox(b) && "sparse solver testing: the rhs should not be modified!");
66     VERIFY(x.isApprox(refX,test_precision<Scalar>()));
67 
68     x.setZero();
69     solve_with_guess(solver, b, x, x);
70     VERIFY(solver.info() == Success && "solving failed when using analyzePattern/factorize API");
71     VERIFY(oldb.isApprox(b) && "sparse solver testing: the rhs should not be modified!");
72     VERIFY(x.isApprox(refX,test_precision<Scalar>()));
73 
74     x.setZero();
75     // test the analyze/factorize API
76     solver.analyzePattern(A);
77     solver.factorize(A);
78     VERIFY(solver.info() == Success && "factorization failed when using analyzePattern/factorize API");
79     x = solver.solve(b);
80     VERIFY(solver.info() == Success && "solving failed when using analyzePattern/factorize API");
81     VERIFY(oldb.isApprox(b) && "sparse solver testing: the rhs should not be modified!");
82     VERIFY(x.isApprox(refX,test_precision<Scalar>()));
83 
84     x.setZero();
85     // test with Map
86     MappedSparseMatrix<Scalar,Mat::Options,StorageIndex> Am(A.rows(), A.cols(), A.nonZeros(), const_cast<StorageIndex*>(A.outerIndexPtr()), const_cast<StorageIndex*>(A.innerIndexPtr()), const_cast<Scalar*>(A.valuePtr()));
87     solver.compute(Am);
88     VERIFY(solver.info() == Success && "factorization failed when using Map");
89     DenseRhs dx(refX);
90     dx.setZero();
91     Map<DenseRhs> xm(dx.data(), dx.rows(), dx.cols());
92     Map<const DenseRhs> bm(db.data(), db.rows(), db.cols());
93     xm = solver.solve(bm);
94     VERIFY(solver.info() == Success && "solving failed when using Map");
95     VERIFY(oldb.isApprox(bm) && "sparse solver testing: the rhs should not be modified!");
96     VERIFY(xm.isApprox(refX,test_precision<Scalar>()));
97   }
98 
99   // if not too large, do some extra check:
100   if(A.rows()<2000)
101   {
102     // test initialization ctor
103     {
104       Rhs x(b.rows(), b.cols());
105       Solver solver2(A);
106       VERIFY(solver2.info() == Success);
107       x = solver2.solve(b);
108       VERIFY(x.isApprox(refX,test_precision<Scalar>()));
109     }
110 
111     // test dense Block as the result and rhs:
112     {
113       DenseRhs x(refX.rows(), refX.cols());
114       DenseRhs oldb(db);
115       x.setZero();
116       x.block(0,0,x.rows(),x.cols()) = solver.solve(db.block(0,0,db.rows(),db.cols()));
117       VERIFY(oldb.isApprox(db) && "sparse solver testing: the rhs should not be modified!");
118       VERIFY(x.isApprox(refX,test_precision<Scalar>()));
119     }
120 
121     // test uncompressed inputs
122     {
123       Mat A2 = A;
124       A2.reserve((ArrayXf::Random(A.outerSize())+2).template cast<typename Mat::StorageIndex>().eval());
125       solver.compute(A2);
126       Rhs x = solver.solve(b);
127       VERIFY(x.isApprox(refX,test_precision<Scalar>()));
128     }
129 
130     // test expression as input
131     {
132       solver.compute(0.5*(A+A));
133       Rhs x = solver.solve(b);
134       VERIFY(x.isApprox(refX,test_precision<Scalar>()));
135 
136       Solver solver2(0.5*(A+A));
137       Rhs x2 = solver2.solve(b);
138       VERIFY(x2.isApprox(refX,test_precision<Scalar>()));
139     }
140   }
141 }
142 
143 template<typename Solver, typename Rhs>
check_sparse_solving_real_cases(Solver & solver,const typename Solver::MatrixType & A,const Rhs & b,const typename Solver::MatrixType & fullA,const Rhs & refX)144 void check_sparse_solving_real_cases(Solver& solver, const typename Solver::MatrixType& A, const Rhs& b, const typename Solver::MatrixType& fullA, const Rhs& refX)
145 {
146   typedef typename Solver::MatrixType Mat;
147   typedef typename Mat::Scalar Scalar;
148   typedef typename Mat::RealScalar RealScalar;
149 
150   Rhs x(A.cols(), b.cols());
151 
152   solver.compute(A);
153   if (solver.info() != Success)
154   {
155     std::cerr << "ERROR | sparse solver testing, factorization failed (" << typeid(Solver).name() << ")\n";
156     VERIFY(solver.info() == Success);
157   }
158   x = solver.solve(b);
159 
160   if (solver.info() != Success)
161   {
162     std::cerr << "WARNING | sparse solver testing, solving failed (" << typeid(Solver).name() << ")\n";
163     return;
164   }
165 
166   RealScalar res_error = (fullA*x-b).norm()/b.norm();
167   VERIFY( (res_error <= test_precision<Scalar>() ) && "sparse solver failed without noticing it");
168 
169 
170   if(refX.size() != 0 && (refX - x).norm()/refX.norm() > test_precision<Scalar>())
171   {
172     std::cerr << "WARNING | found solution is different from the provided reference one\n";
173   }
174 
175 }
176 template<typename Solver, typename DenseMat>
check_sparse_determinant(Solver & solver,const typename Solver::MatrixType & A,const DenseMat & dA)177 void check_sparse_determinant(Solver& solver, const typename Solver::MatrixType& A, const DenseMat& dA)
178 {
179   typedef typename Solver::MatrixType Mat;
180   typedef typename Mat::Scalar Scalar;
181 
182   solver.compute(A);
183   if (solver.info() != Success)
184   {
185     std::cerr << "WARNING | sparse solver testing: factorization failed (check_sparse_determinant)\n";
186     return;
187   }
188 
189   Scalar refDet = dA.determinant();
190   VERIFY_IS_APPROX(refDet,solver.determinant());
191 }
192 template<typename Solver, typename DenseMat>
check_sparse_abs_determinant(Solver & solver,const typename Solver::MatrixType & A,const DenseMat & dA)193 void check_sparse_abs_determinant(Solver& solver, const typename Solver::MatrixType& A, const DenseMat& dA)
194 {
195   using std::abs;
196   typedef typename Solver::MatrixType Mat;
197   typedef typename Mat::Scalar Scalar;
198 
199   solver.compute(A);
200   if (solver.info() != Success)
201   {
202     std::cerr << "WARNING | sparse solver testing: factorization failed (check_sparse_abs_determinant)\n";
203     return;
204   }
205 
206   Scalar refDet = abs(dA.determinant());
207   VERIFY_IS_APPROX(refDet,solver.absDeterminant());
208 }
209 
210 template<typename Solver, typename DenseMat>
211 int generate_sparse_spd_problem(Solver& , typename Solver::MatrixType& A, typename Solver::MatrixType& halfA, DenseMat& dA, int maxSize = 300)
212 {
213   typedef typename Solver::MatrixType Mat;
214   typedef typename Mat::Scalar Scalar;
215   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
216 
217   int size = internal::random<int>(1,maxSize);
218   double density = (std::max)(8./(size*size), 0.01);
219 
220   Mat M(size, size);
221   DenseMatrix dM(size, size);
222 
223   initSparse<Scalar>(density, dM, M, ForceNonZeroDiag);
224 
225   A = M * M.adjoint();
226   dA = dM * dM.adjoint();
227 
228   halfA.resize(size,size);
229   if(Solver::UpLo==(Lower|Upper))
230     halfA = A;
231   else
232     halfA.template selfadjointView<Solver::UpLo>().rankUpdate(M);
233 
234   return size;
235 }
236 
237 
238 #ifdef TEST_REAL_CASES
239 template<typename Scalar>
get_matrixfolder()240 inline std::string get_matrixfolder()
241 {
242   std::string mat_folder = TEST_REAL_CASES;
243   if( internal::is_same<Scalar, std::complex<float> >::value || internal::is_same<Scalar, std::complex<double> >::value )
244     mat_folder  = mat_folder + static_cast<std::string>("/complex/");
245   else
246     mat_folder = mat_folder + static_cast<std::string>("/real/");
247   return mat_folder;
248 }
sym_to_string(int sym)249 std::string sym_to_string(int sym)
250 {
251   if(sym==Symmetric) return "Symmetric ";
252   if(sym==SPD)       return "SPD ";
253   return "";
254 }
255 template<typename Derived>
solver_stats(const IterativeSolverBase<Derived> & solver)256 std::string solver_stats(const IterativeSolverBase<Derived> &solver)
257 {
258   std::stringstream ss;
259   ss << solver.iterations() << " iters, error: " << solver.error();
260   return ss.str();
261 }
262 template<typename Derived>
solver_stats(const SparseSolverBase<Derived> &)263 std::string solver_stats(const SparseSolverBase<Derived> &/*solver*/)
264 {
265   return "";
266 }
267 #endif
268 
269 template<typename Solver> void check_sparse_spd_solving(Solver& solver, int maxSize = 300, int maxRealWorldSize = 100000)
270 {
271   typedef typename Solver::MatrixType Mat;
272   typedef typename Mat::Scalar Scalar;
273   typedef typename Mat::StorageIndex StorageIndex;
274   typedef SparseMatrix<Scalar,ColMajor, StorageIndex> SpMat;
275   typedef SparseVector<Scalar, 0, StorageIndex> SpVec;
276   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
277   typedef Matrix<Scalar,Dynamic,1> DenseVector;
278 
279   // generate the problem
280   Mat A, halfA;
281   DenseMatrix dA;
282   for (int i = 0; i < g_repeat; i++) {
283     int size = generate_sparse_spd_problem(solver, A, halfA, dA, maxSize);
284 
285     // generate the right hand sides
286     int rhsCols = internal::random<int>(1,16);
287     double density = (std::max)(8./(size*rhsCols), 0.1);
288     SpMat B(size,rhsCols);
289     DenseVector b = DenseVector::Random(size);
290     DenseMatrix dB(size,rhsCols);
291     initSparse<Scalar>(density, dB, B, ForceNonZeroDiag);
292     SpVec c = B.col(0);
293     DenseVector dc = dB.col(0);
294 
295     CALL_SUBTEST( check_sparse_solving(solver, A,     b,  dA, b)  );
296     CALL_SUBTEST( check_sparse_solving(solver, halfA, b,  dA, b)  );
297     CALL_SUBTEST( check_sparse_solving(solver, A,     dB, dA, dB) );
298     CALL_SUBTEST( check_sparse_solving(solver, halfA, dB, dA, dB) );
299     CALL_SUBTEST( check_sparse_solving(solver, A,     B,  dA, dB) );
300     CALL_SUBTEST( check_sparse_solving(solver, halfA, B,  dA, dB) );
301     CALL_SUBTEST( check_sparse_solving(solver, A,     c,  dA, dc) );
302     CALL_SUBTEST( check_sparse_solving(solver, halfA, c,  dA, dc) );
303 
304     // check only once
305     if(i==0)
306     {
307       b = DenseVector::Zero(size);
308       check_sparse_solving(solver, A, b, dA, b);
309     }
310   }
311 
312   // First, get the folder
313 #ifdef TEST_REAL_CASES
314   // Test real problems with double precision only
315   if (internal::is_same<typename NumTraits<Scalar>::Real, double>::value)
316   {
317     std::string mat_folder = get_matrixfolder<Scalar>();
318     MatrixMarketIterator<Scalar> it(mat_folder);
319     for (; it; ++it)
320     {
321       if (it.sym() == SPD){
322         A = it.matrix();
323         if(A.diagonal().size() <= maxRealWorldSize)
324         {
325           DenseVector b = it.rhs();
326           DenseVector refX = it.refX();
327           PermutationMatrix<Dynamic, Dynamic, StorageIndex> pnull;
328           halfA.resize(A.rows(), A.cols());
329           if(Solver::UpLo == (Lower|Upper))
330             halfA = A;
331           else
332             halfA.template selfadjointView<Solver::UpLo>() = A.template triangularView<Eigen::Lower>().twistedBy(pnull);
333 
334           std::cout << "INFO | Testing " << sym_to_string(it.sym()) << "sparse problem " << it.matname()
335                   << " (" << A.rows() << "x" << A.cols() << ") using " << typeid(Solver).name() << "..." << std::endl;
336           CALL_SUBTEST( check_sparse_solving_real_cases(solver, A,     b, A, refX) );
337           std::string stats = solver_stats(solver);
338           if(stats.size()>0)
339             std::cout << "INFO |  " << stats << std::endl;
340           CALL_SUBTEST( check_sparse_solving_real_cases(solver, halfA, b, A, refX) );
341         }
342         else
343         {
344           std::cout << "INFO | Skip sparse problem \"" << it.matname() << "\" (too large)" << std::endl;
345         }
346       }
347     }
348   }
349 #else
350   EIGEN_UNUSED_VARIABLE(maxRealWorldSize);
351 #endif
352 }
353 
check_sparse_spd_determinant(Solver & solver)354 template<typename Solver> void check_sparse_spd_determinant(Solver& solver)
355 {
356   typedef typename Solver::MatrixType Mat;
357   typedef typename Mat::Scalar Scalar;
358   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
359 
360   // generate the problem
361   Mat A, halfA;
362   DenseMatrix dA;
363   generate_sparse_spd_problem(solver, A, halfA, dA, 30);
364 
365   for (int i = 0; i < g_repeat; i++) {
366     check_sparse_determinant(solver, A,     dA);
367     check_sparse_determinant(solver, halfA, dA );
368   }
369 }
370 
371 template<typename Solver, typename DenseMat>
372 Index generate_sparse_square_problem(Solver&, typename Solver::MatrixType& A, DenseMat& dA, int maxSize = 300, int options = ForceNonZeroDiag)
373 {
374   typedef typename Solver::MatrixType Mat;
375   typedef typename Mat::Scalar Scalar;
376 
377   Index size = internal::random<int>(1,maxSize);
378   double density = (std::max)(8./(size*size), 0.01);
379 
380   A.resize(size,size);
381   dA.resize(size,size);
382 
383   initSparse<Scalar>(density, dA, A, options);
384 
385   return size;
386 }
387 
388 
389 struct prune_column {
390   Index m_col;
prune_columnprune_column391   prune_column(Index col) : m_col(col) {}
392   template<class Scalar>
operatorprune_column393   bool operator()(Index, Index col, const Scalar&) const {
394     return col != m_col;
395   }
396 };
397 
398 
399 template<typename Solver> void check_sparse_square_solving(Solver& solver, int maxSize = 300, int maxRealWorldSize = 100000, bool checkDeficient = false)
400 {
401   typedef typename Solver::MatrixType Mat;
402   typedef typename Mat::Scalar Scalar;
403   typedef SparseMatrix<Scalar,ColMajor, typename Mat::StorageIndex> SpMat;
404   typedef SparseVector<Scalar, 0, typename Mat::StorageIndex> SpVec;
405   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
406   typedef Matrix<Scalar,Dynamic,1> DenseVector;
407 
408   int rhsCols = internal::random<int>(1,16);
409 
410   Mat A;
411   DenseMatrix dA;
412   for (int i = 0; i < g_repeat; i++) {
413     Index size = generate_sparse_square_problem(solver, A, dA, maxSize);
414 
415     A.makeCompressed();
416     DenseVector b = DenseVector::Random(size);
417     DenseMatrix dB(size,rhsCols);
418     SpMat B(size,rhsCols);
419     double density = (std::max)(8./(size*rhsCols), 0.1);
420     initSparse<Scalar>(density, dB, B, ForceNonZeroDiag);
421     B.makeCompressed();
422     SpVec c = B.col(0);
423     DenseVector dc = dB.col(0);
424     CALL_SUBTEST(check_sparse_solving(solver, A, b,  dA, b));
425     CALL_SUBTEST(check_sparse_solving(solver, A, dB, dA, dB));
426     CALL_SUBTEST(check_sparse_solving(solver, A, B,  dA, dB));
427     CALL_SUBTEST(check_sparse_solving(solver, A, c,  dA, dc));
428 
429     // check only once
430     if(i==0)
431     {
432       b = DenseVector::Zero(size);
433       check_sparse_solving(solver, A, b, dA, b);
434     }
435     // regression test for Bug 792 (structurally rank deficient matrices):
436     if(checkDeficient && size>1) {
437       Index col = internal::random<int>(0,int(size-1));
438       A.prune(prune_column(col));
439       solver.compute(A);
440       VERIFY_IS_EQUAL(solver.info(), NumericalIssue);
441     }
442   }
443 
444   // First, get the folder
445 #ifdef TEST_REAL_CASES
446   // Test real problems with double precision only
447   if (internal::is_same<typename NumTraits<Scalar>::Real, double>::value)
448   {
449     std::string mat_folder = get_matrixfolder<Scalar>();
450     MatrixMarketIterator<Scalar> it(mat_folder);
451     for (; it; ++it)
452     {
453       A = it.matrix();
454       if(A.diagonal().size() <= maxRealWorldSize)
455       {
456         DenseVector b = it.rhs();
457         DenseVector refX = it.refX();
458         std::cout << "INFO | Testing " << sym_to_string(it.sym()) << "sparse problem " << it.matname()
459                   << " (" << A.rows() << "x" << A.cols() << ") using " << typeid(Solver).name() << "..." << std::endl;
460         CALL_SUBTEST(check_sparse_solving_real_cases(solver, A, b, A, refX));
461         std::string stats = solver_stats(solver);
462         if(stats.size()>0)
463           std::cout << "INFO |  " << stats << std::endl;
464       }
465       else
466       {
467         std::cout << "INFO | SKIP sparse problem \"" << it.matname() << "\" (too large)" << std::endl;
468       }
469     }
470   }
471 #else
472   EIGEN_UNUSED_VARIABLE(maxRealWorldSize);
473 #endif
474 
475 }
476 
check_sparse_square_determinant(Solver & solver)477 template<typename Solver> void check_sparse_square_determinant(Solver& solver)
478 {
479   typedef typename Solver::MatrixType Mat;
480   typedef typename Mat::Scalar Scalar;
481   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
482 
483   for (int i = 0; i < g_repeat; i++) {
484     // generate the problem
485     Mat A;
486     DenseMatrix dA;
487 
488     int size = internal::random<int>(1,30);
489     dA.setRandom(size,size);
490 
491     dA = (dA.array().abs()<0.3).select(0,dA);
492     dA.diagonal() = (dA.diagonal().array()==0).select(1,dA.diagonal());
493     A = dA.sparseView();
494     A.makeCompressed();
495 
496     check_sparse_determinant(solver, A, dA);
497   }
498 }
499 
check_sparse_square_abs_determinant(Solver & solver)500 template<typename Solver> void check_sparse_square_abs_determinant(Solver& solver)
501 {
502   typedef typename Solver::MatrixType Mat;
503   typedef typename Mat::Scalar Scalar;
504   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
505 
506   for (int i = 0; i < g_repeat; i++) {
507     // generate the problem
508     Mat A;
509     DenseMatrix dA;
510     generate_sparse_square_problem(solver, A, dA, 30);
511     A.makeCompressed();
512     check_sparse_abs_determinant(solver, A, dA);
513   }
514 }
515 
516 template<typename Solver, typename DenseMat>
517 void generate_sparse_leastsquare_problem(Solver&, typename Solver::MatrixType& A, DenseMat& dA, int maxSize = 300, int options = ForceNonZeroDiag)
518 {
519   typedef typename Solver::MatrixType Mat;
520   typedef typename Mat::Scalar Scalar;
521 
522   int rows = internal::random<int>(1,maxSize);
523   int cols = internal::random<int>(1,rows);
524   double density = (std::max)(8./(rows*cols), 0.01);
525 
526   A.resize(rows,cols);
527   dA.resize(rows,cols);
528 
529   initSparse<Scalar>(density, dA, A, options);
530 }
531 
check_sparse_leastsquare_solving(Solver & solver)532 template<typename Solver> void check_sparse_leastsquare_solving(Solver& solver)
533 {
534   typedef typename Solver::MatrixType Mat;
535   typedef typename Mat::Scalar Scalar;
536   typedef SparseMatrix<Scalar,ColMajor, typename Mat::StorageIndex> SpMat;
537   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
538   typedef Matrix<Scalar,Dynamic,1> DenseVector;
539 
540   int rhsCols = internal::random<int>(1,16);
541 
542   Mat A;
543   DenseMatrix dA;
544   for (int i = 0; i < g_repeat; i++) {
545     generate_sparse_leastsquare_problem(solver, A, dA);
546 
547     A.makeCompressed();
548     DenseVector b = DenseVector::Random(A.rows());
549     DenseMatrix dB(A.rows(),rhsCols);
550     SpMat B(A.rows(),rhsCols);
551     double density = (std::max)(8./(A.rows()*rhsCols), 0.1);
552     initSparse<Scalar>(density, dB, B, ForceNonZeroDiag);
553     B.makeCompressed();
554     check_sparse_solving(solver, A, b,  dA, b);
555     check_sparse_solving(solver, A, dB, dA, dB);
556     check_sparse_solving(solver, A, B,  dA, dB);
557 
558     // check only once
559     if(i==0)
560     {
561       b = DenseVector::Zero(A.rows());
562       check_sparse_solving(solver, A, b, dA, b);
563     }
564   }
565 }
566