• 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) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
5 // Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
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 SVD_DEFAULT
12 #error a macro SVD_DEFAULT(MatrixType) must be defined prior to including svd_common.h
13 #endif
14 
15 #ifndef SVD_FOR_MIN_NORM
16 #error a macro SVD_FOR_MIN_NORM(MatrixType) must be defined prior to including svd_common.h
17 #endif
18 
19 #include "svd_fill.h"
20 
21 // Check that the matrix m is properly reconstructed and that the U and V factors are unitary
22 // The SVD must have already been computed.
23 template<typename SvdType, typename MatrixType>
svd_check_full(const MatrixType & m,const SvdType & svd)24 void svd_check_full(const MatrixType& m, const SvdType& svd)
25 {
26   typedef typename MatrixType::Index Index;
27   Index rows = m.rows();
28   Index cols = m.cols();
29 
30   enum {
31     RowsAtCompileTime = MatrixType::RowsAtCompileTime,
32     ColsAtCompileTime = MatrixType::ColsAtCompileTime
33   };
34 
35   typedef typename MatrixType::Scalar Scalar;
36   typedef typename MatrixType::RealScalar RealScalar;
37   typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime> MatrixUType;
38   typedef Matrix<Scalar, ColsAtCompileTime, ColsAtCompileTime> MatrixVType;
39 
40   MatrixType sigma = MatrixType::Zero(rows,cols);
41   sigma.diagonal() = svd.singularValues().template cast<Scalar>();
42   MatrixUType u = svd.matrixU();
43   MatrixVType v = svd.matrixV();
44   RealScalar scaling = m.cwiseAbs().maxCoeff();
45   if(scaling<(std::numeric_limits<RealScalar>::min)())
46   {
47     VERIFY(sigma.cwiseAbs().maxCoeff() <= (std::numeric_limits<RealScalar>::min)());
48   }
49   else
50   {
51     VERIFY_IS_APPROX(m/scaling, u * (sigma/scaling) * v.adjoint());
52   }
53   VERIFY_IS_UNITARY(u);
54   VERIFY_IS_UNITARY(v);
55 }
56 
57 // Compare partial SVD defined by computationOptions to a full SVD referenceSvd
58 template<typename SvdType, typename MatrixType>
svd_compare_to_full(const MatrixType & m,unsigned int computationOptions,const SvdType & referenceSvd)59 void svd_compare_to_full(const MatrixType& m,
60                          unsigned int computationOptions,
61                          const SvdType& referenceSvd)
62 {
63   typedef typename MatrixType::RealScalar RealScalar;
64   Index rows = m.rows();
65   Index cols = m.cols();
66   Index diagSize = (std::min)(rows, cols);
67   RealScalar prec = test_precision<RealScalar>();
68 
69   SvdType svd(m, computationOptions);
70 
71   VERIFY_IS_APPROX(svd.singularValues(), referenceSvd.singularValues());
72 
73   if(computationOptions & (ComputeFullV|ComputeThinV))
74   {
75     VERIFY( (svd.matrixV().adjoint()*svd.matrixV()).isIdentity(prec) );
76     VERIFY_IS_APPROX( svd.matrixV().leftCols(diagSize) * svd.singularValues().asDiagonal() * svd.matrixV().leftCols(diagSize).adjoint(),
77                       referenceSvd.matrixV().leftCols(diagSize) * referenceSvd.singularValues().asDiagonal() * referenceSvd.matrixV().leftCols(diagSize).adjoint());
78   }
79 
80   if(computationOptions & (ComputeFullU|ComputeThinU))
81   {
82     VERIFY( (svd.matrixU().adjoint()*svd.matrixU()).isIdentity(prec) );
83     VERIFY_IS_APPROX( svd.matrixU().leftCols(diagSize) * svd.singularValues().cwiseAbs2().asDiagonal() * svd.matrixU().leftCols(diagSize).adjoint(),
84                       referenceSvd.matrixU().leftCols(diagSize) * referenceSvd.singularValues().cwiseAbs2().asDiagonal() * referenceSvd.matrixU().leftCols(diagSize).adjoint());
85   }
86 
87   // The following checks are not critical.
88   // For instance, with Dived&Conquer SVD, if only the factor 'V' is computedt then different matrix-matrix product implementation will be used
89   // and the resulting 'V' factor might be significantly different when the SVD decomposition is not unique, especially with single precision float.
90   ++g_test_level;
91   if(computationOptions & ComputeFullU)  VERIFY_IS_APPROX(svd.matrixU(), referenceSvd.matrixU());
92   if(computationOptions & ComputeThinU)  VERIFY_IS_APPROX(svd.matrixU(), referenceSvd.matrixU().leftCols(diagSize));
93   if(computationOptions & ComputeFullV)  VERIFY_IS_APPROX(svd.matrixV().cwiseAbs(), referenceSvd.matrixV().cwiseAbs());
94   if(computationOptions & ComputeThinV)  VERIFY_IS_APPROX(svd.matrixV(), referenceSvd.matrixV().leftCols(diagSize));
95   --g_test_level;
96 }
97 
98 //
99 template<typename SvdType, typename MatrixType>
svd_least_square(const MatrixType & m,unsigned int computationOptions)100 void svd_least_square(const MatrixType& m, unsigned int computationOptions)
101 {
102   typedef typename MatrixType::Scalar Scalar;
103   typedef typename MatrixType::RealScalar RealScalar;
104   typedef typename MatrixType::Index Index;
105   Index rows = m.rows();
106   Index cols = m.cols();
107 
108   enum {
109     RowsAtCompileTime = MatrixType::RowsAtCompileTime,
110     ColsAtCompileTime = MatrixType::ColsAtCompileTime
111   };
112 
113   typedef Matrix<Scalar, RowsAtCompileTime, Dynamic> RhsType;
114   typedef Matrix<Scalar, ColsAtCompileTime, Dynamic> SolutionType;
115 
116   RhsType rhs = RhsType::Random(rows, internal::random<Index>(1, cols));
117   SvdType svd(m, computationOptions);
118 
119        if(internal::is_same<RealScalar,double>::value) svd.setThreshold(1e-8);
120   else if(internal::is_same<RealScalar,float>::value)  svd.setThreshold(2e-4);
121 
122   SolutionType x = svd.solve(rhs);
123 
124   RealScalar residual = (m*x-rhs).norm();
125   RealScalar rhs_norm = rhs.norm();
126   if(!test_isMuchSmallerThan(residual,rhs.norm()))
127   {
128     // ^^^ If the residual is very small, then we have an exact solution, so we are already good.
129 
130     // evaluate normal equation which works also for least-squares solutions
131     if(internal::is_same<RealScalar,double>::value || svd.rank()==m.diagonal().size())
132     {
133       using std::sqrt;
134       // This test is not stable with single precision.
135       // This is probably because squaring m signicantly affects the precision.
136       if(internal::is_same<RealScalar,float>::value) ++g_test_level;
137 
138       VERIFY_IS_APPROX(m.adjoint()*(m*x),m.adjoint()*rhs);
139 
140       if(internal::is_same<RealScalar,float>::value) --g_test_level;
141     }
142 
143     // Check that there is no significantly better solution in the neighborhood of x
144     for(Index k=0;k<x.rows();++k)
145     {
146       using std::abs;
147 
148       SolutionType y(x);
149       y.row(k) = (RealScalar(1)+2*NumTraits<RealScalar>::epsilon())*x.row(k);
150       RealScalar residual_y = (m*y-rhs).norm();
151       VERIFY( test_isMuchSmallerThan(abs(residual_y-residual), rhs_norm) || residual < residual_y );
152       if(internal::is_same<RealScalar,float>::value) ++g_test_level;
153       VERIFY( test_isApprox(residual_y,residual) || residual < residual_y );
154       if(internal::is_same<RealScalar,float>::value) --g_test_level;
155 
156       y.row(k) = (RealScalar(1)-2*NumTraits<RealScalar>::epsilon())*x.row(k);
157       residual_y = (m*y-rhs).norm();
158       VERIFY( test_isMuchSmallerThan(abs(residual_y-residual), rhs_norm) || residual < residual_y );
159       if(internal::is_same<RealScalar,float>::value) ++g_test_level;
160       VERIFY( test_isApprox(residual_y,residual) || residual < residual_y );
161       if(internal::is_same<RealScalar,float>::value) --g_test_level;
162     }
163   }
164 }
165 
166 // check minimal norm solutions, the inoput matrix m is only used to recover problem size
167 template<typename MatrixType>
svd_min_norm(const MatrixType & m,unsigned int computationOptions)168 void svd_min_norm(const MatrixType& m, unsigned int computationOptions)
169 {
170   typedef typename MatrixType::Scalar Scalar;
171   typedef typename MatrixType::Index Index;
172   Index cols = m.cols();
173 
174   enum {
175     ColsAtCompileTime = MatrixType::ColsAtCompileTime
176   };
177 
178   typedef Matrix<Scalar, ColsAtCompileTime, Dynamic> SolutionType;
179 
180   // generate a full-rank m x n problem with m<n
181   enum {
182     RankAtCompileTime2 = ColsAtCompileTime==Dynamic ? Dynamic : (ColsAtCompileTime)/2+1,
183     RowsAtCompileTime3 = ColsAtCompileTime==Dynamic ? Dynamic : ColsAtCompileTime+1
184   };
185   typedef Matrix<Scalar, RankAtCompileTime2, ColsAtCompileTime> MatrixType2;
186   typedef Matrix<Scalar, RankAtCompileTime2, 1> RhsType2;
187   typedef Matrix<Scalar, ColsAtCompileTime, RankAtCompileTime2> MatrixType2T;
188   Index rank = RankAtCompileTime2==Dynamic ? internal::random<Index>(1,cols) : Index(RankAtCompileTime2);
189   MatrixType2 m2(rank,cols);
190   int guard = 0;
191   do {
192     m2.setRandom();
193   } while(SVD_FOR_MIN_NORM(MatrixType2)(m2).setThreshold(test_precision<Scalar>()).rank()!=rank && (++guard)<10);
194   VERIFY(guard<10);
195 
196   RhsType2 rhs2 = RhsType2::Random(rank);
197   // use QR to find a reference minimal norm solution
198   HouseholderQR<MatrixType2T> qr(m2.adjoint());
199   Matrix<Scalar,Dynamic,1> tmp = qr.matrixQR().topLeftCorner(rank,rank).template triangularView<Upper>().adjoint().solve(rhs2);
200   tmp.conservativeResize(cols);
201   tmp.tail(cols-rank).setZero();
202   SolutionType x21 = qr.householderQ() * tmp;
203   // now check with SVD
204   SVD_FOR_MIN_NORM(MatrixType2) svd2(m2, computationOptions);
205   SolutionType x22 = svd2.solve(rhs2);
206   VERIFY_IS_APPROX(m2*x21, rhs2);
207   VERIFY_IS_APPROX(m2*x22, rhs2);
208   VERIFY_IS_APPROX(x21, x22);
209 
210   // Now check with a rank deficient matrix
211   typedef Matrix<Scalar, RowsAtCompileTime3, ColsAtCompileTime> MatrixType3;
212   typedef Matrix<Scalar, RowsAtCompileTime3, 1> RhsType3;
213   Index rows3 = RowsAtCompileTime3==Dynamic ? internal::random<Index>(rank+1,2*cols) : Index(RowsAtCompileTime3);
214   Matrix<Scalar,RowsAtCompileTime3,Dynamic> C = Matrix<Scalar,RowsAtCompileTime3,Dynamic>::Random(rows3,rank);
215   MatrixType3 m3 = C * m2;
216   RhsType3 rhs3 = C * rhs2;
217   SVD_FOR_MIN_NORM(MatrixType3) svd3(m3, computationOptions);
218   SolutionType x3 = svd3.solve(rhs3);
219   VERIFY_IS_APPROX(m3*x3, rhs3);
220   VERIFY_IS_APPROX(m3*x21, rhs3);
221   VERIFY_IS_APPROX(m2*x3, rhs2);
222   VERIFY_IS_APPROX(x21, x3);
223 }
224 
225 // Check full, compare_to_full, least_square, and min_norm for all possible compute-options
226 template<typename SvdType, typename MatrixType>
svd_test_all_computation_options(const MatrixType & m,bool full_only)227 void svd_test_all_computation_options(const MatrixType& m, bool full_only)
228 {
229 //   if (QRPreconditioner == NoQRPreconditioner && m.rows() != m.cols())
230 //     return;
231   SvdType fullSvd(m, ComputeFullU|ComputeFullV);
232   CALL_SUBTEST(( svd_check_full(m, fullSvd) ));
233   CALL_SUBTEST(( svd_least_square<SvdType>(m, ComputeFullU | ComputeFullV) ));
234   CALL_SUBTEST(( svd_min_norm(m, ComputeFullU | ComputeFullV) ));
235 
236   #if defined __INTEL_COMPILER
237   // remark #111: statement is unreachable
238   #pragma warning disable 111
239   #endif
240   if(full_only)
241     return;
242 
243   CALL_SUBTEST(( svd_compare_to_full(m, ComputeFullU, fullSvd) ));
244   CALL_SUBTEST(( svd_compare_to_full(m, ComputeFullV, fullSvd) ));
245   CALL_SUBTEST(( svd_compare_to_full(m, 0, fullSvd) ));
246 
247   if (MatrixType::ColsAtCompileTime == Dynamic) {
248     // thin U/V are only available with dynamic number of columns
249     CALL_SUBTEST(( svd_compare_to_full(m, ComputeFullU|ComputeThinV, fullSvd) ));
250     CALL_SUBTEST(( svd_compare_to_full(m,              ComputeThinV, fullSvd) ));
251     CALL_SUBTEST(( svd_compare_to_full(m, ComputeThinU|ComputeFullV, fullSvd) ));
252     CALL_SUBTEST(( svd_compare_to_full(m, ComputeThinU             , fullSvd) ));
253     CALL_SUBTEST(( svd_compare_to_full(m, ComputeThinU|ComputeThinV, fullSvd) ));
254 
255     CALL_SUBTEST(( svd_least_square<SvdType>(m, ComputeFullU | ComputeThinV) ));
256     CALL_SUBTEST(( svd_least_square<SvdType>(m, ComputeThinU | ComputeFullV) ));
257     CALL_SUBTEST(( svd_least_square<SvdType>(m, ComputeThinU | ComputeThinV) ));
258 
259     CALL_SUBTEST(( svd_min_norm(m, ComputeFullU | ComputeThinV) ));
260     CALL_SUBTEST(( svd_min_norm(m, ComputeThinU | ComputeFullV) ));
261     CALL_SUBTEST(( svd_min_norm(m, ComputeThinU | ComputeThinV) ));
262 
263     // test reconstruction
264     typedef typename MatrixType::Index Index;
265     Index diagSize = (std::min)(m.rows(), m.cols());
266     SvdType svd(m, ComputeThinU | ComputeThinV);
267     VERIFY_IS_APPROX(m, svd.matrixU().leftCols(diagSize) * svd.singularValues().asDiagonal() * svd.matrixV().leftCols(diagSize).adjoint());
268   }
269 }
270 
271 
272 // work around stupid msvc error when constructing at compile time an expression that involves
273 // a division by zero, even if the numeric type has floating point
274 template<typename Scalar>
zero()275 EIGEN_DONT_INLINE Scalar zero() { return Scalar(0); }
276 
277 // workaround aggressive optimization in ICC
sub(T a,T b)278 template<typename T> EIGEN_DONT_INLINE  T sub(T a, T b) { return a - b; }
279 
280 // all this function does is verify we don't iterate infinitely on nan/inf values
281 template<typename SvdType, typename MatrixType>
svd_inf_nan()282 void svd_inf_nan()
283 {
284   SvdType svd;
285   typedef typename MatrixType::Scalar Scalar;
286   Scalar some_inf = Scalar(1) / zero<Scalar>();
287   VERIFY(sub(some_inf, some_inf) != sub(some_inf, some_inf));
288   svd.compute(MatrixType::Constant(10,10,some_inf), ComputeFullU | ComputeFullV);
289 
290   Scalar nan = std::numeric_limits<Scalar>::quiet_NaN();
291   VERIFY(nan != nan);
292   svd.compute(MatrixType::Constant(10,10,nan), ComputeFullU | ComputeFullV);
293 
294   MatrixType m = MatrixType::Zero(10,10);
295   m(internal::random<int>(0,9), internal::random<int>(0,9)) = some_inf;
296   svd.compute(m, ComputeFullU | ComputeFullV);
297 
298   m = MatrixType::Zero(10,10);
299   m(internal::random<int>(0,9), internal::random<int>(0,9)) = nan;
300   svd.compute(m, ComputeFullU | ComputeFullV);
301 
302   // regression test for bug 791
303   m.resize(3,3);
304   m << 0,    2*NumTraits<Scalar>::epsilon(),  0.5,
305        0,   -0.5,                             0,
306        nan,  0,                               0;
307   svd.compute(m, ComputeFullU | ComputeFullV);
308 
309   m.resize(4,4);
310   m <<  1, 0, 0, 0,
311         0, 3, 1, 2e-308,
312         1, 0, 1, nan,
313         0, nan, nan, 0;
314   svd.compute(m, ComputeFullU | ComputeFullV);
315 }
316 
317 // Regression test for bug 286: JacobiSVD loops indefinitely with some
318 // matrices containing denormal numbers.
319 template<typename>
svd_underoverflow()320 void svd_underoverflow()
321 {
322 #if defined __INTEL_COMPILER
323 // shut up warning #239: floating point underflow
324 #pragma warning push
325 #pragma warning disable 239
326 #endif
327   Matrix2d M;
328   M << -7.90884e-313, -4.94e-324,
329                  0, 5.60844e-313;
330   SVD_DEFAULT(Matrix2d) svd;
331   svd.compute(M,ComputeFullU|ComputeFullV);
332   CALL_SUBTEST( svd_check_full(M,svd) );
333 
334   // Check all 2x2 matrices made with the following coefficients:
335   VectorXd value_set(9);
336   value_set << 0, 1, -1, 5.60844e-313, -5.60844e-313, 4.94e-324, -4.94e-324, -4.94e-223, 4.94e-223;
337   Array4i id(0,0,0,0);
338   int k = 0;
339   do
340   {
341     M << value_set(id(0)), value_set(id(1)), value_set(id(2)), value_set(id(3));
342     svd.compute(M,ComputeFullU|ComputeFullV);
343     CALL_SUBTEST( svd_check_full(M,svd) );
344 
345     id(k)++;
346     if(id(k)>=value_set.size())
347     {
348       while(k<3 && id(k)>=value_set.size()) id(++k)++;
349       id.head(k).setZero();
350       k=0;
351     }
352 
353   } while((id<int(value_set.size())).all());
354 
355 #if defined __INTEL_COMPILER
356 #pragma warning pop
357 #endif
358 
359   // Check for overflow:
360   Matrix3d M3;
361   M3 << 4.4331978442502944e+307, -5.8585363752028680e+307,  6.4527017443412964e+307,
362         3.7841695601406358e+307,  2.4331702789740617e+306, -3.5235707140272905e+307,
363        -8.7190887618028355e+307, -7.3453213709232193e+307, -2.4367363684472105e+307;
364 
365   SVD_DEFAULT(Matrix3d) svd3;
366   svd3.compute(M3,ComputeFullU|ComputeFullV); // just check we don't loop indefinitely
367   CALL_SUBTEST( svd_check_full(M3,svd3) );
368 }
369 
370 // void jacobisvd(const MatrixType& a = MatrixType(), bool pickrandom = true)
371 
372 template<typename MatrixType>
svd_all_trivial_2x2(void (* cb)(const MatrixType &,bool))373 void svd_all_trivial_2x2( void (*cb)(const MatrixType&,bool) )
374 {
375   MatrixType M;
376   VectorXd value_set(3);
377   value_set << 0, 1, -1;
378   Array4i id(0,0,0,0);
379   int k = 0;
380   do
381   {
382     M << value_set(id(0)), value_set(id(1)), value_set(id(2)), value_set(id(3));
383 
384     cb(M,false);
385 
386     id(k)++;
387     if(id(k)>=value_set.size())
388     {
389       while(k<3 && id(k)>=value_set.size()) id(++k)++;
390       id.head(k).setZero();
391       k=0;
392     }
393 
394   } while((id<int(value_set.size())).all());
395 }
396 
397 template<typename>
svd_preallocate()398 void svd_preallocate()
399 {
400   Vector3f v(3.f, 2.f, 1.f);
401   MatrixXf m = v.asDiagonal();
402 
403   internal::set_is_malloc_allowed(false);
404   VERIFY_RAISES_ASSERT(VectorXf tmp(10);)
405   SVD_DEFAULT(MatrixXf) svd;
406   internal::set_is_malloc_allowed(true);
407   svd.compute(m);
408   VERIFY_IS_APPROX(svd.singularValues(), v);
409 
410   SVD_DEFAULT(MatrixXf) svd2(3,3);
411   internal::set_is_malloc_allowed(false);
412   svd2.compute(m);
413   internal::set_is_malloc_allowed(true);
414   VERIFY_IS_APPROX(svd2.singularValues(), v);
415   VERIFY_RAISES_ASSERT(svd2.matrixU());
416   VERIFY_RAISES_ASSERT(svd2.matrixV());
417   svd2.compute(m, ComputeFullU | ComputeFullV);
418   VERIFY_IS_APPROX(svd2.matrixU(), Matrix3f::Identity());
419   VERIFY_IS_APPROX(svd2.matrixV(), Matrix3f::Identity());
420   internal::set_is_malloc_allowed(false);
421   svd2.compute(m);
422   internal::set_is_malloc_allowed(true);
423 
424   SVD_DEFAULT(MatrixXf) svd3(3,3,ComputeFullU|ComputeFullV);
425   internal::set_is_malloc_allowed(false);
426   svd2.compute(m);
427   internal::set_is_malloc_allowed(true);
428   VERIFY_IS_APPROX(svd2.singularValues(), v);
429   VERIFY_IS_APPROX(svd2.matrixU(), Matrix3f::Identity());
430   VERIFY_IS_APPROX(svd2.matrixV(), Matrix3f::Identity());
431   internal::set_is_malloc_allowed(false);
432   svd2.compute(m, ComputeFullU|ComputeFullV);
433   internal::set_is_malloc_allowed(true);
434 }
435 
436 template<typename SvdType,typename MatrixType>
svd_verify_assert(const MatrixType & m)437 void svd_verify_assert(const MatrixType& m)
438 {
439   typedef typename MatrixType::Scalar Scalar;
440   typedef typename MatrixType::Index Index;
441   Index rows = m.rows();
442   Index cols = m.cols();
443 
444   enum {
445     RowsAtCompileTime = MatrixType::RowsAtCompileTime,
446     ColsAtCompileTime = MatrixType::ColsAtCompileTime
447   };
448 
449   typedef Matrix<Scalar, RowsAtCompileTime, 1> RhsType;
450   RhsType rhs(rows);
451   SvdType svd;
452   VERIFY_RAISES_ASSERT(svd.matrixU())
453   VERIFY_RAISES_ASSERT(svd.singularValues())
454   VERIFY_RAISES_ASSERT(svd.matrixV())
455   VERIFY_RAISES_ASSERT(svd.solve(rhs))
456   MatrixType a = MatrixType::Zero(rows, cols);
457   a.setZero();
458   svd.compute(a, 0);
459   VERIFY_RAISES_ASSERT(svd.matrixU())
460   VERIFY_RAISES_ASSERT(svd.matrixV())
461   svd.singularValues();
462   VERIFY_RAISES_ASSERT(svd.solve(rhs))
463 
464   if (ColsAtCompileTime == Dynamic)
465   {
466     svd.compute(a, ComputeThinU);
467     svd.matrixU();
468     VERIFY_RAISES_ASSERT(svd.matrixV())
469     VERIFY_RAISES_ASSERT(svd.solve(rhs))
470     svd.compute(a, ComputeThinV);
471     svd.matrixV();
472     VERIFY_RAISES_ASSERT(svd.matrixU())
473     VERIFY_RAISES_ASSERT(svd.solve(rhs))
474   }
475   else
476   {
477     VERIFY_RAISES_ASSERT(svd.compute(a, ComputeThinU))
478     VERIFY_RAISES_ASSERT(svd.compute(a, ComputeThinV))
479   }
480 }
481 
482 #undef SVD_DEFAULT
483 #undef SVD_FOR_MIN_NORM
484