• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 #include <iostream>
2 #include <Eigen/Core>
3 #include <Eigen/Dense>
4 #include <Eigen/IterativeLinearSolvers>
5 #include <unsupported/Eigen/IterativeSolvers>
6 
7 class MatrixReplacement;
8 using Eigen::SparseMatrix;
9 
10 namespace Eigen {
11 namespace internal {
12   // MatrixReplacement looks-like a SparseMatrix, so let's inherits its traits:
13   template<>
14   struct traits<MatrixReplacement> :  public Eigen::internal::traits<Eigen::SparseMatrix<double> >
15   {};
16 }
17 }
18 
19 // Example of a matrix-free wrapper from a user type to Eigen's compatible type
20 // For the sake of simplicity, this example simply wrap a Eigen::SparseMatrix.
21 class MatrixReplacement : public Eigen::EigenBase<MatrixReplacement> {
22 public:
23   // Required typedefs, constants, and method:
24   typedef double Scalar;
25   typedef double RealScalar;
26   typedef int StorageIndex;
27   enum {
28     ColsAtCompileTime = Eigen::Dynamic,
29     MaxColsAtCompileTime = Eigen::Dynamic,
30     IsRowMajor = false
31   };
32 
rows() const33   Index rows() const { return mp_mat->rows(); }
cols() const34   Index cols() const { return mp_mat->cols(); }
35 
36   template<typename Rhs>
operator *(const Eigen::MatrixBase<Rhs> & x) const37   Eigen::Product<MatrixReplacement,Rhs,Eigen::AliasFreeProduct> operator*(const Eigen::MatrixBase<Rhs>& x) const {
38     return Eigen::Product<MatrixReplacement,Rhs,Eigen::AliasFreeProduct>(*this, x.derived());
39   }
40 
41   // Custom API:
MatrixReplacement()42   MatrixReplacement() : mp_mat(0) {}
43 
attachMyMatrix(const SparseMatrix<double> & mat)44   void attachMyMatrix(const SparseMatrix<double> &mat) {
45     mp_mat = &mat;
46   }
my_matrix() const47   const SparseMatrix<double> my_matrix() const { return *mp_mat; }
48 
49 private:
50   const SparseMatrix<double> *mp_mat;
51 };
52 
53 
54 // Implementation of MatrixReplacement * Eigen::DenseVector though a specialization of internal::generic_product_impl:
55 namespace Eigen {
56 namespace internal {
57 
58   template<typename Rhs>
59   struct generic_product_impl<MatrixReplacement, Rhs, SparseShape, DenseShape, GemvProduct> // GEMV stands for matrix-vector
60   : generic_product_impl_base<MatrixReplacement,Rhs,generic_product_impl<MatrixReplacement,Rhs> >
61   {
62     typedef typename Product<MatrixReplacement,Rhs>::Scalar Scalar;
63 
64     template<typename Dest>
scaleAndAddToEigen::internal::generic_product_impl65     static void scaleAndAddTo(Dest& dst, const MatrixReplacement& lhs, const Rhs& rhs, const Scalar& alpha)
66     {
67       // This method should implement "dst += alpha * lhs * rhs" inplace,
68       // however, for iterative solvers, alpha is always equal to 1, so let's not bother about it.
69       assert(alpha==Scalar(1) && "scaling is not implemented");
70 
71       // Here we could simply call dst.noalias() += lhs.my_matrix() * rhs,
72       // but let's do something fancier (and less efficient):
73       for(Index i=0; i<lhs.cols(); ++i)
74         dst += rhs(i) * lhs.my_matrix().col(i);
75     }
76   };
77 
78 }
79 }
80 
main()81 int main()
82 {
83   int n = 10;
84   Eigen::SparseMatrix<double> S = Eigen::MatrixXd::Random(n,n).sparseView(0.5,1);
85   S = S.transpose()*S;
86 
87   MatrixReplacement A;
88   A.attachMyMatrix(S);
89 
90   Eigen::VectorXd b(n), x;
91   b.setRandom();
92 
93   // Solve Ax = b using various iterative solver with matrix-free version:
94   {
95     Eigen::ConjugateGradient<MatrixReplacement, Eigen::Lower|Eigen::Upper, Eigen::IdentityPreconditioner> cg;
96     cg.compute(A);
97     x = cg.solve(b);
98     std::cout << "CG:       #iterations: " << cg.iterations() << ", estimated error: " << cg.error() << std::endl;
99   }
100 
101   {
102     Eigen::BiCGSTAB<MatrixReplacement, Eigen::IdentityPreconditioner> bicg;
103     bicg.compute(A);
104     x = bicg.solve(b);
105     std::cout << "BiCGSTAB: #iterations: " << bicg.iterations() << ", estimated error: " << bicg.error() << std::endl;
106   }
107 
108   {
109     Eigen::GMRES<MatrixReplacement, Eigen::IdentityPreconditioner> gmres;
110     gmres.compute(A);
111     x = gmres.solve(b);
112     std::cout << "GMRES:    #iterations: " << gmres.iterations() << ", estimated error: " << gmres.error() << std::endl;
113   }
114 
115   {
116     Eigen::DGMRES<MatrixReplacement, Eigen::IdentityPreconditioner> gmres;
117     gmres.compute(A);
118     x = gmres.solve(b);
119     std::cout << "DGMRES:   #iterations: " << gmres.iterations() << ", estimated error: " << gmres.error() << std::endl;
120   }
121 
122   {
123     Eigen::MINRES<MatrixReplacement, Eigen::Lower|Eigen::Upper, Eigen::IdentityPreconditioner> minres;
124     minres.compute(A);
125     x = minres.solve(b);
126     std::cout << "MINRES:   #iterations: " << minres.iterations() << ", estimated error: " << minres.error() << std::endl;
127   }
128 }
129