• 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 <Eigen/SparseLU>
13 #include <sstream>
14 
15 template<typename Solver, typename Rhs, typename Guess,typename Result>
solve_with_guess(IterativeSolverBase<Solver> & solver,const MatrixBase<Rhs> & b,const Guess & g,Result & x)16 void solve_with_guess(IterativeSolverBase<Solver>& solver, const MatrixBase<Rhs>& b, const Guess& g, Result &x) {
17   if(internal::random<bool>())
18   {
19     // With a temporary through evaluator<SolveWithGuess>
20     x = solver.derived().solveWithGuess(b,g) + Result::Zero(x.rows(), x.cols());
21   }
22   else
23   {
24     // direct evaluation within x through Assignment<Result,SolveWithGuess>
25     x = solver.derived().solveWithGuess(b.derived(),g);
26   }
27 }
28 
29 template<typename Solver, typename Rhs, typename Guess,typename Result>
solve_with_guess(SparseSolverBase<Solver> & solver,const MatrixBase<Rhs> & b,const Guess &,Result & x)30 void solve_with_guess(SparseSolverBase<Solver>& solver, const MatrixBase<Rhs>& b, const Guess& , Result& x) {
31   if(internal::random<bool>())
32     x = solver.derived().solve(b) + Result::Zero(x.rows(), x.cols());
33   else
34     x = solver.derived().solve(b);
35 }
36 
37 template<typename Solver, typename Rhs, typename Guess,typename Result>
solve_with_guess(SparseSolverBase<Solver> & solver,const SparseMatrixBase<Rhs> & b,const Guess &,Result & x)38 void solve_with_guess(SparseSolverBase<Solver>& solver, const SparseMatrixBase<Rhs>& b, const Guess& , Result& x) {
39   x = solver.derived().solve(b);
40 }
41 
42 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)43 void check_sparse_solving(Solver& solver, const typename Solver::MatrixType& A, const Rhs& b, const DenseMat& dA, const DenseRhs& db)
44 {
45   typedef typename Solver::MatrixType Mat;
46   typedef typename Mat::Scalar Scalar;
47   typedef typename Mat::StorageIndex StorageIndex;
48 
49   DenseRhs refX = dA.householderQr().solve(db);
50   {
51     Rhs x(A.cols(), b.cols());
52     Rhs oldb = b;
53 
54     solver.compute(A);
55     if (solver.info() != Success)
56     {
57       std::cerr << "ERROR | sparse solver testing, factorization failed (" << typeid(Solver).name() << ")\n";
58       VERIFY(solver.info() == Success);
59     }
60     x = solver.solve(b);
61     if (solver.info() != Success)
62     {
63       std::cerr << "WARNING: sparse solver testing: solving failed (" << typeid(Solver).name() << ")\n";
64       // dump call stack:
65       g_test_level++;
66       VERIFY(solver.info() == Success);
67       g_test_level--;
68       return;
69     }
70     VERIFY(oldb.isApprox(b) && "sparse solver testing: the rhs should not be modified!");
71     VERIFY(x.isApprox(refX,test_precision<Scalar>()));
72 
73     x.setZero();
74     solve_with_guess(solver, b, x, x);
75     VERIFY(solver.info() == Success && "solving failed when using solve_with_guess API");
76     VERIFY(oldb.isApprox(b) && "sparse solver testing: the rhs should not be modified!");
77     VERIFY(x.isApprox(refX,test_precision<Scalar>()));
78 
79     x.setZero();
80     // test the analyze/factorize API
81     solver.analyzePattern(A);
82     solver.factorize(A);
83     VERIFY(solver.info() == Success && "factorization failed when using analyzePattern/factorize API");
84     x = solver.solve(b);
85     VERIFY(solver.info() == Success && "solving failed when using analyzePattern/factorize API");
86     VERIFY(oldb.isApprox(b) && "sparse solver testing: the rhs should not be modified!");
87     VERIFY(x.isApprox(refX,test_precision<Scalar>()));
88 
89     x.setZero();
90     // test with Map
91     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()));
92     solver.compute(Am);
93     VERIFY(solver.info() == Success && "factorization failed when using Map");
94     DenseRhs dx(refX);
95     dx.setZero();
96     Map<DenseRhs> xm(dx.data(), dx.rows(), dx.cols());
97     Map<const DenseRhs> bm(db.data(), db.rows(), db.cols());
98     xm = solver.solve(bm);
99     VERIFY(solver.info() == Success && "solving failed when using Map");
100     VERIFY(oldb.isApprox(bm) && "sparse solver testing: the rhs should not be modified!");
101     VERIFY(xm.isApprox(refX,test_precision<Scalar>()));
102   }
103 
104   // if not too large, do some extra check:
105   if(A.rows()<2000)
106   {
107     // test initialization ctor
108     {
109       Rhs x(b.rows(), b.cols());
110       Solver solver2(A);
111       VERIFY(solver2.info() == Success);
112       x = solver2.solve(b);
113       VERIFY(x.isApprox(refX,test_precision<Scalar>()));
114     }
115 
116     // test dense Block as the result and rhs:
117     {
118       DenseRhs x(refX.rows(), refX.cols());
119       DenseRhs oldb(db);
120       x.setZero();
121       x.block(0,0,x.rows(),x.cols()) = solver.solve(db.block(0,0,db.rows(),db.cols()));
122       VERIFY(oldb.isApprox(db) && "sparse solver testing: the rhs should not be modified!");
123       VERIFY(x.isApprox(refX,test_precision<Scalar>()));
124     }
125 
126     // test uncompressed inputs
127     {
128       Mat A2 = A;
129       A2.reserve((ArrayXf::Random(A.outerSize())+2).template cast<typename Mat::StorageIndex>().eval());
130       solver.compute(A2);
131       Rhs x = solver.solve(b);
132       VERIFY(x.isApprox(refX,test_precision<Scalar>()));
133     }
134 
135     // test expression as input
136     {
137       solver.compute(0.5*(A+A));
138       Rhs x = solver.solve(b);
139       VERIFY(x.isApprox(refX,test_precision<Scalar>()));
140 
141       Solver solver2(0.5*(A+A));
142       Rhs x2 = solver2.solve(b);
143       VERIFY(x2.isApprox(refX,test_precision<Scalar>()));
144     }
145   }
146 }
147 
148 // specialization of generic check_sparse_solving for SuperLU in order to also test adjoint and transpose solves
149 template<typename Scalar, typename Rhs, typename DenseMat, typename DenseRhs>
check_sparse_solving(Eigen::SparseLU<Eigen::SparseMatrix<Scalar>> & solver,const typename Eigen::SparseMatrix<Scalar> & A,const Rhs & b,const DenseMat & dA,const DenseRhs & db)150 void check_sparse_solving(Eigen::SparseLU<Eigen::SparseMatrix<Scalar> >& solver, const typename Eigen::SparseMatrix<Scalar>& A, const Rhs& b, const DenseMat& dA, const DenseRhs& db)
151 {
152   typedef typename Eigen::SparseMatrix<Scalar> Mat;
153   typedef typename Mat::StorageIndex StorageIndex;
154   typedef typename Eigen::SparseLU<Eigen::SparseMatrix<Scalar> > Solver;
155 
156   // reference solutions computed by dense QR solver
157   DenseRhs refX1 = dA.householderQr().solve(db); // solution of A x = db
158   DenseRhs refX2 = dA.transpose().householderQr().solve(db); // solution of A^T * x = db (use transposed matrix A^T)
159   DenseRhs refX3 = dA.adjoint().householderQr().solve(db);  // solution of A^* * x = db (use adjoint matrix A^*)
160 
161 
162   {
163     Rhs x1(A.cols(), b.cols());
164     Rhs x2(A.cols(), b.cols());
165     Rhs x3(A.cols(), b.cols());
166     Rhs oldb = b;
167 
168     solver.compute(A);
169     if (solver.info() != Success)
170     {
171       std::cerr << "ERROR | sparse solver testing, factorization failed (" << typeid(Solver).name() << ")\n";
172       VERIFY(solver.info() == Success);
173     }
174     x1 = solver.solve(b);
175     if (solver.info() != Success)
176     {
177       std::cerr << "WARNING | sparse solver testing: solving failed (" << typeid(Solver).name() << ")\n";
178       return;
179     }
180     VERIFY(oldb.isApprox(b,0.0) && "sparse solver testing: the rhs should not be modified!");
181     VERIFY(x1.isApprox(refX1,test_precision<Scalar>()));
182 
183     // test solve with transposed
184     x2 = solver.transpose().solve(b);
185     VERIFY(oldb.isApprox(b) && "sparse solver testing: the rhs should not be modified!");
186     VERIFY(x2.isApprox(refX2,test_precision<Scalar>()));
187 
188 
189     // test solve with adjoint
190     //solver.template _solve_impl_transposed<true>(b, x3);
191     x3 = solver.adjoint().solve(b);
192     VERIFY(oldb.isApprox(b,0.0) && "sparse solver testing: the rhs should not be modified!");
193     VERIFY(x3.isApprox(refX3,test_precision<Scalar>()));
194 
195     x1.setZero();
196     solve_with_guess(solver, b, x1, x1);
197     VERIFY(solver.info() == Success && "solving failed when using analyzePattern/factorize API");
198     VERIFY(oldb.isApprox(b,0.0) && "sparse solver testing: the rhs should not be modified!");
199     VERIFY(x1.isApprox(refX1,test_precision<Scalar>()));
200 
201     x1.setZero();
202     x2.setZero();
203     x3.setZero();
204     // test the analyze/factorize API
205     solver.analyzePattern(A);
206     solver.factorize(A);
207     VERIFY(solver.info() == Success && "factorization failed when using analyzePattern/factorize API");
208     x1 = solver.solve(b);
209     x2 = solver.transpose().solve(b);
210     x3 = solver.adjoint().solve(b);
211 
212     VERIFY(solver.info() == Success && "solving failed when using analyzePattern/factorize API");
213     VERIFY(oldb.isApprox(b,0.0) && "sparse solver testing: the rhs should not be modified!");
214     VERIFY(x1.isApprox(refX1,test_precision<Scalar>()));
215     VERIFY(x2.isApprox(refX2,test_precision<Scalar>()));
216     VERIFY(x3.isApprox(refX3,test_precision<Scalar>()));
217 
218     x1.setZero();
219     // test with Map
220     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()));
221     solver.compute(Am);
222     VERIFY(solver.info() == Success && "factorization failed when using Map");
223     DenseRhs dx(refX1);
224     dx.setZero();
225     Map<DenseRhs> xm(dx.data(), dx.rows(), dx.cols());
226     Map<const DenseRhs> bm(db.data(), db.rows(), db.cols());
227     xm = solver.solve(bm);
228     VERIFY(solver.info() == Success && "solving failed when using Map");
229     VERIFY(oldb.isApprox(bm,0.0) && "sparse solver testing: the rhs should not be modified!");
230     VERIFY(xm.isApprox(refX1,test_precision<Scalar>()));
231   }
232 
233   // if not too large, do some extra check:
234   if(A.rows()<2000)
235   {
236     // test initialization ctor
237     {
238       Rhs x(b.rows(), b.cols());
239       Solver solver2(A);
240       VERIFY(solver2.info() == Success);
241       x = solver2.solve(b);
242       VERIFY(x.isApprox(refX1,test_precision<Scalar>()));
243     }
244 
245     // test dense Block as the result and rhs:
246     {
247       DenseRhs x(refX1.rows(), refX1.cols());
248       DenseRhs oldb(db);
249       x.setZero();
250       x.block(0,0,x.rows(),x.cols()) = solver.solve(db.block(0,0,db.rows(),db.cols()));
251       VERIFY(oldb.isApprox(db,0.0) && "sparse solver testing: the rhs should not be modified!");
252       VERIFY(x.isApprox(refX1,test_precision<Scalar>()));
253     }
254 
255     // test uncompressed inputs
256     {
257       Mat A2 = A;
258       A2.reserve((ArrayXf::Random(A.outerSize())+2).template cast<typename Mat::StorageIndex>().eval());
259       solver.compute(A2);
260       Rhs x = solver.solve(b);
261       VERIFY(x.isApprox(refX1,test_precision<Scalar>()));
262     }
263 
264     // test expression as input
265     {
266       solver.compute(0.5*(A+A));
267       Rhs x = solver.solve(b);
268       VERIFY(x.isApprox(refX1,test_precision<Scalar>()));
269 
270       Solver solver2(0.5*(A+A));
271       Rhs x2 = solver2.solve(b);
272       VERIFY(x2.isApprox(refX1,test_precision<Scalar>()));
273     }
274   }
275 }
276 
277 
278 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)279 void check_sparse_solving_real_cases(Solver& solver, const typename Solver::MatrixType& A, const Rhs& b, const typename Solver::MatrixType& fullA, const Rhs& refX)
280 {
281   typedef typename Solver::MatrixType Mat;
282   typedef typename Mat::Scalar Scalar;
283   typedef typename Mat::RealScalar RealScalar;
284 
285   Rhs x(A.cols(), b.cols());
286 
287   solver.compute(A);
288   if (solver.info() != Success)
289   {
290     std::cerr << "ERROR | sparse solver testing, factorization failed (" << typeid(Solver).name() << ")\n";
291     VERIFY(solver.info() == Success);
292   }
293   x = solver.solve(b);
294 
295   if (solver.info() != Success)
296   {
297     std::cerr << "WARNING | sparse solver testing, solving failed (" << typeid(Solver).name() << ")\n";
298     return;
299   }
300 
301   RealScalar res_error = (fullA*x-b).norm()/b.norm();
302   VERIFY( (res_error <= test_precision<Scalar>() ) && "sparse solver failed without noticing it");
303 
304 
305   if(refX.size() != 0 && (refX - x).norm()/refX.norm() > test_precision<Scalar>())
306   {
307     std::cerr << "WARNING | found solution is different from the provided reference one\n";
308   }
309 
310 }
311 template<typename Solver, typename DenseMat>
check_sparse_determinant(Solver & solver,const typename Solver::MatrixType & A,const DenseMat & dA)312 void check_sparse_determinant(Solver& solver, const typename Solver::MatrixType& A, const DenseMat& dA)
313 {
314   typedef typename Solver::MatrixType Mat;
315   typedef typename Mat::Scalar Scalar;
316 
317   solver.compute(A);
318   if (solver.info() != Success)
319   {
320     std::cerr << "WARNING | sparse solver testing: factorization failed (check_sparse_determinant)\n";
321     return;
322   }
323 
324   Scalar refDet = dA.determinant();
325   VERIFY_IS_APPROX(refDet,solver.determinant());
326 }
327 template<typename Solver, typename DenseMat>
check_sparse_abs_determinant(Solver & solver,const typename Solver::MatrixType & A,const DenseMat & dA)328 void check_sparse_abs_determinant(Solver& solver, const typename Solver::MatrixType& A, const DenseMat& dA)
329 {
330   using std::abs;
331   typedef typename Solver::MatrixType Mat;
332   typedef typename Mat::Scalar Scalar;
333 
334   solver.compute(A);
335   if (solver.info() != Success)
336   {
337     std::cerr << "WARNING | sparse solver testing: factorization failed (check_sparse_abs_determinant)\n";
338     return;
339   }
340 
341   Scalar refDet = abs(dA.determinant());
342   VERIFY_IS_APPROX(refDet,solver.absDeterminant());
343 }
344 
345 template<typename Solver, typename DenseMat>
346 int generate_sparse_spd_problem(Solver& , typename Solver::MatrixType& A, typename Solver::MatrixType& halfA, DenseMat& dA, int maxSize = 300)
347 {
348   typedef typename Solver::MatrixType Mat;
349   typedef typename Mat::Scalar Scalar;
350   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
351 
352   int size = internal::random<int>(1,maxSize);
353   double density = (std::max)(8./(size*size), 0.01);
354 
355   Mat M(size, size);
356   DenseMatrix dM(size, size);
357 
358   initSparse<Scalar>(density, dM, M, ForceNonZeroDiag);
359 
360   A = M * M.adjoint();
361   dA = dM * dM.adjoint();
362 
363   halfA.resize(size,size);
364   if(Solver::UpLo==(Lower|Upper))
365     halfA = A;
366   else
367     halfA.template selfadjointView<Solver::UpLo>().rankUpdate(M);
368 
369   return size;
370 }
371 
372 
373 #ifdef TEST_REAL_CASES
374 template<typename Scalar>
get_matrixfolder()375 inline std::string get_matrixfolder()
376 {
377   std::string mat_folder = TEST_REAL_CASES;
378   if( internal::is_same<Scalar, std::complex<float> >::value || internal::is_same<Scalar, std::complex<double> >::value )
379     mat_folder  = mat_folder + static_cast<std::string>("/complex/");
380   else
381     mat_folder = mat_folder + static_cast<std::string>("/real/");
382   return mat_folder;
383 }
sym_to_string(int sym)384 std::string sym_to_string(int sym)
385 {
386   if(sym==Symmetric) return "Symmetric ";
387   if(sym==SPD)       return "SPD ";
388   return "";
389 }
390 template<typename Derived>
solver_stats(const IterativeSolverBase<Derived> & solver)391 std::string solver_stats(const IterativeSolverBase<Derived> &solver)
392 {
393   std::stringstream ss;
394   ss << solver.iterations() << " iters, error: " << solver.error();
395   return ss.str();
396 }
397 template<typename Derived>
solver_stats(const SparseSolverBase<Derived> &)398 std::string solver_stats(const SparseSolverBase<Derived> &/*solver*/)
399 {
400   return "";
401 }
402 #endif
403 
404 template<typename Solver> void check_sparse_spd_solving(Solver& solver, int maxSize = (std::min)(300,EIGEN_TEST_MAX_SIZE), int maxRealWorldSize = 100000)
405 {
406   typedef typename Solver::MatrixType Mat;
407   typedef typename Mat::Scalar Scalar;
408   typedef typename Mat::StorageIndex StorageIndex;
409   typedef SparseMatrix<Scalar,ColMajor, StorageIndex> SpMat;
410   typedef SparseVector<Scalar, 0, StorageIndex> SpVec;
411   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
412   typedef Matrix<Scalar,Dynamic,1> DenseVector;
413 
414   // generate the problem
415   Mat A, halfA;
416   DenseMatrix dA;
417   for (int i = 0; i < g_repeat; i++) {
418     int size = generate_sparse_spd_problem(solver, A, halfA, dA, maxSize);
419 
420     // generate the right hand sides
421     int rhsCols = internal::random<int>(1,16);
422     double density = (std::max)(8./(size*rhsCols), 0.1);
423     SpMat B(size,rhsCols);
424     DenseVector b = DenseVector::Random(size);
425     DenseMatrix dB(size,rhsCols);
426     initSparse<Scalar>(density, dB, B, ForceNonZeroDiag);
427     SpVec c = B.col(0);
428     DenseVector dc = dB.col(0);
429 
430     CALL_SUBTEST( check_sparse_solving(solver, A,     b,  dA, b)  );
431     CALL_SUBTEST( check_sparse_solving(solver, halfA, b,  dA, b)  );
432     CALL_SUBTEST( check_sparse_solving(solver, A,     dB, dA, dB) );
433     CALL_SUBTEST( check_sparse_solving(solver, halfA, dB, dA, dB) );
434     CALL_SUBTEST( check_sparse_solving(solver, A,     B,  dA, dB) );
435     CALL_SUBTEST( check_sparse_solving(solver, halfA, B,  dA, dB) );
436     CALL_SUBTEST( check_sparse_solving(solver, A,     c,  dA, dc) );
437     CALL_SUBTEST( check_sparse_solving(solver, halfA, c,  dA, dc) );
438 
439     // check only once
440     if(i==0)
441     {
442       b = DenseVector::Zero(size);
443       check_sparse_solving(solver, A, b, dA, b);
444     }
445   }
446 
447   // First, get the folder
448 #ifdef TEST_REAL_CASES
449   // Test real problems with double precision only
450   if (internal::is_same<typename NumTraits<Scalar>::Real, double>::value)
451   {
452     std::string mat_folder = get_matrixfolder<Scalar>();
453     MatrixMarketIterator<Scalar> it(mat_folder);
454     for (; it; ++it)
455     {
456       if (it.sym() == SPD){
457         A = it.matrix();
458         if(A.diagonal().size() <= maxRealWorldSize)
459         {
460           DenseVector b = it.rhs();
461           DenseVector refX = it.refX();
462           PermutationMatrix<Dynamic, Dynamic, StorageIndex> pnull;
463           halfA.resize(A.rows(), A.cols());
464           if(Solver::UpLo == (Lower|Upper))
465             halfA = A;
466           else
467             halfA.template selfadjointView<Solver::UpLo>() = A.template triangularView<Eigen::Lower>().twistedBy(pnull);
468 
469           std::cout << "INFO | Testing " << sym_to_string(it.sym()) << "sparse problem " << it.matname()
470                   << " (" << A.rows() << "x" << A.cols() << ") using " << typeid(Solver).name() << "..." << std::endl;
471           CALL_SUBTEST( check_sparse_solving_real_cases(solver, A,     b, A, refX) );
472           std::string stats = solver_stats(solver);
473           if(stats.size()>0)
474             std::cout << "INFO |  " << stats << std::endl;
475           CALL_SUBTEST( check_sparse_solving_real_cases(solver, halfA, b, A, refX) );
476         }
477         else
478         {
479           std::cout << "INFO | Skip sparse problem \"" << it.matname() << "\" (too large)" << std::endl;
480         }
481       }
482     }
483   }
484 #else
485   EIGEN_UNUSED_VARIABLE(maxRealWorldSize);
486 #endif
487 }
488 
check_sparse_spd_determinant(Solver & solver)489 template<typename Solver> void check_sparse_spd_determinant(Solver& solver)
490 {
491   typedef typename Solver::MatrixType Mat;
492   typedef typename Mat::Scalar Scalar;
493   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
494 
495   // generate the problem
496   Mat A, halfA;
497   DenseMatrix dA;
498   generate_sparse_spd_problem(solver, A, halfA, dA, 30);
499 
500   for (int i = 0; i < g_repeat; i++) {
501     check_sparse_determinant(solver, A,     dA);
502     check_sparse_determinant(solver, halfA, dA );
503   }
504 }
505 
506 template<typename Solver, typename DenseMat>
507 Index generate_sparse_square_problem(Solver&, typename Solver::MatrixType& A, DenseMat& dA, int maxSize = 300, int options = ForceNonZeroDiag)
508 {
509   typedef typename Solver::MatrixType Mat;
510   typedef typename Mat::Scalar Scalar;
511 
512   Index size = internal::random<int>(1,maxSize);
513   double density = (std::max)(8./(size*size), 0.01);
514 
515   A.resize(size,size);
516   dA.resize(size,size);
517 
518   initSparse<Scalar>(density, dA, A, options);
519 
520   return size;
521 }
522 
523 
524 struct prune_column {
525   Index m_col;
prune_columnprune_column526   prune_column(Index col) : m_col(col) {}
527   template<class Scalar>
operatorprune_column528   bool operator()(Index, Index col, const Scalar&) const {
529     return col != m_col;
530   }
531 };
532 
533 
534 template<typename Solver> void check_sparse_square_solving(Solver& solver, int maxSize = 300, int maxRealWorldSize = 100000, bool checkDeficient = false)
535 {
536   typedef typename Solver::MatrixType Mat;
537   typedef typename Mat::Scalar Scalar;
538   typedef SparseMatrix<Scalar,ColMajor, typename Mat::StorageIndex> SpMat;
539   typedef SparseVector<Scalar, 0, typename Mat::StorageIndex> SpVec;
540   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
541   typedef Matrix<Scalar,Dynamic,1> DenseVector;
542 
543   int rhsCols = internal::random<int>(1,16);
544 
545   Mat A;
546   DenseMatrix dA;
547   for (int i = 0; i < g_repeat; i++) {
548     Index size = generate_sparse_square_problem(solver, A, dA, maxSize);
549 
550     A.makeCompressed();
551     DenseVector b = DenseVector::Random(size);
552     DenseMatrix dB(size,rhsCols);
553     SpMat B(size,rhsCols);
554     double density = (std::max)(8./(size*rhsCols), 0.1);
555     initSparse<Scalar>(density, dB, B, ForceNonZeroDiag);
556     B.makeCompressed();
557     SpVec c = B.col(0);
558     DenseVector dc = dB.col(0);
559     CALL_SUBTEST(check_sparse_solving(solver, A, b,  dA, b));
560     CALL_SUBTEST(check_sparse_solving(solver, A, dB, dA, dB));
561     CALL_SUBTEST(check_sparse_solving(solver, A, B,  dA, dB));
562     CALL_SUBTEST(check_sparse_solving(solver, A, c,  dA, dc));
563 
564     // check only once
565     if(i==0)
566     {
567       CALL_SUBTEST(b = DenseVector::Zero(size); check_sparse_solving(solver, A, b, dA, b));
568     }
569     // regression test for Bug 792 (structurally rank deficient matrices):
570     if(checkDeficient && size>1) {
571       Index col = internal::random<int>(0,int(size-1));
572       A.prune(prune_column(col));
573       solver.compute(A);
574       VERIFY_IS_EQUAL(solver.info(), NumericalIssue);
575     }
576   }
577 
578   // First, get the folder
579 #ifdef TEST_REAL_CASES
580   // Test real problems with double precision only
581   if (internal::is_same<typename NumTraits<Scalar>::Real, double>::value)
582   {
583     std::string mat_folder = get_matrixfolder<Scalar>();
584     MatrixMarketIterator<Scalar> it(mat_folder);
585     for (; it; ++it)
586     {
587       A = it.matrix();
588       if(A.diagonal().size() <= maxRealWorldSize)
589       {
590         DenseVector b = it.rhs();
591         DenseVector refX = it.refX();
592         std::cout << "INFO | Testing " << sym_to_string(it.sym()) << "sparse problem " << it.matname()
593                   << " (" << A.rows() << "x" << A.cols() << ") using " << typeid(Solver).name() << "..." << std::endl;
594         CALL_SUBTEST(check_sparse_solving_real_cases(solver, A, b, A, refX));
595         std::string stats = solver_stats(solver);
596         if(stats.size()>0)
597           std::cout << "INFO |  " << stats << std::endl;
598       }
599       else
600       {
601         std::cout << "INFO | SKIP sparse problem \"" << it.matname() << "\" (too large)" << std::endl;
602       }
603     }
604   }
605 #else
606   EIGEN_UNUSED_VARIABLE(maxRealWorldSize);
607 #endif
608 
609 }
610 
check_sparse_square_determinant(Solver & solver)611 template<typename Solver> void check_sparse_square_determinant(Solver& solver)
612 {
613   typedef typename Solver::MatrixType Mat;
614   typedef typename Mat::Scalar Scalar;
615   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
616 
617   for (int i = 0; i < g_repeat; i++) {
618     // generate the problem
619     Mat A;
620     DenseMatrix dA;
621 
622     int size = internal::random<int>(1,30);
623     dA.setRandom(size,size);
624 
625     dA = (dA.array().abs()<0.3).select(0,dA);
626     dA.diagonal() = (dA.diagonal().array()==0).select(1,dA.diagonal());
627     A = dA.sparseView();
628     A.makeCompressed();
629 
630     check_sparse_determinant(solver, A, dA);
631   }
632 }
633 
check_sparse_square_abs_determinant(Solver & solver)634 template<typename Solver> void check_sparse_square_abs_determinant(Solver& solver)
635 {
636   typedef typename Solver::MatrixType Mat;
637   typedef typename Mat::Scalar Scalar;
638   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
639 
640   for (int i = 0; i < g_repeat; i++) {
641     // generate the problem
642     Mat A;
643     DenseMatrix dA;
644     generate_sparse_square_problem(solver, A, dA, 30);
645     A.makeCompressed();
646     check_sparse_abs_determinant(solver, A, dA);
647   }
648 }
649 
650 template<typename Solver, typename DenseMat>
651 void generate_sparse_leastsquare_problem(Solver&, typename Solver::MatrixType& A, DenseMat& dA, int maxSize = 300, int options = ForceNonZeroDiag)
652 {
653   typedef typename Solver::MatrixType Mat;
654   typedef typename Mat::Scalar Scalar;
655 
656   int rows = internal::random<int>(1,maxSize);
657   int cols = internal::random<int>(1,rows);
658   double density = (std::max)(8./(rows*cols), 0.01);
659 
660   A.resize(rows,cols);
661   dA.resize(rows,cols);
662 
663   initSparse<Scalar>(density, dA, A, options);
664 }
665 
check_sparse_leastsquare_solving(Solver & solver)666 template<typename Solver> void check_sparse_leastsquare_solving(Solver& solver)
667 {
668   typedef typename Solver::MatrixType Mat;
669   typedef typename Mat::Scalar Scalar;
670   typedef SparseMatrix<Scalar,ColMajor, typename Mat::StorageIndex> SpMat;
671   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
672   typedef Matrix<Scalar,Dynamic,1> DenseVector;
673 
674   int rhsCols = internal::random<int>(1,16);
675 
676   Mat A;
677   DenseMatrix dA;
678   for (int i = 0; i < g_repeat; i++) {
679     generate_sparse_leastsquare_problem(solver, A, dA);
680 
681     A.makeCompressed();
682     DenseVector b = DenseVector::Random(A.rows());
683     DenseMatrix dB(A.rows(),rhsCols);
684     SpMat B(A.rows(),rhsCols);
685     double density = (std::max)(8./(A.rows()*rhsCols), 0.1);
686     initSparse<Scalar>(density, dB, B, ForceNonZeroDiag);
687     B.makeCompressed();
688     check_sparse_solving(solver, A, b,  dA, b);
689     check_sparse_solving(solver, A, dB, dA, dB);
690     check_sparse_solving(solver, A, B,  dA, dB);
691 
692     // check only once
693     if(i==0)
694     {
695       b = DenseVector::Zero(A.rows());
696       check_sparse_solving(solver, A, b, dA, b);
697     }
698   }
699 }
700