Hi, developers!
I was used Eigen in my project.
I have Sparse Matrix data in CRS or CCS format like next structre

Code: Select all

struct MySpMatr
{
    int iNNZ - number of non-zero elements
    int *ia - pointer like Eigen::SparseMatrix::m_outerIndex
    int  *ja - pointer like CompressedStorage::m_indices
    type  *a - pointer like CompressedStorage::m_values
    int rows - number of matrix rows
   int cols - number of matrix columns
}

I woudn't use new memory like use SparseMatrix::setFromTriplets or
SparseMatrix::resize(newsize1, newsize2);
StorageIndex* pIn = SparseMatrix::outerIndexPtr();

Code: Select all

for (int ii=0; ii<newsize1; ii++)
{
     pIn[ii] = ia[ii];
}
Storage Stor = SparseMatrix::data();
Stor.clear();
Stor.resize(  /// etc

I want use my stored data in Eigen::SparseMatrix
In my project was created litle fake wrapper, this code

Code: Select all

namespace Eigen {
namespace internal {
// This fake class with access to protected members of  CompressedStorage
template<typename _Scalar, typename _StorageIndex>
class CompressedStorage2 : public CompressedStorage <_Scalar, _StorageIndex>
{
 public:
       CompressedStorage2():CompressedStorage(){};
   void setExternalData(_StorageIndex* innerIndex, _Scalar* values, int iNNZ)
   {
      m_size = iNNZ;
      m_values = values;
      m_indices = innerIndex;
      m_allocatedSize = iNNZ;
   }
   void freeExternalData()
   {
      m_size = 0;
      m_values = 0;
      m_indices = 0;
      m_allocatedSize = 0;
   }

};
}
}


namespace Eigen {
template<typename _Scalar, int _Options, typename _Index>
class SparseMatrix2 : public SparseMatrix<_Scalar, _Options, _Index>
{
public:
   typedef internal::CompressedStorage2<Scalar,StorageIndex> Storage2;
   inline SparseMatrix2():SparseMatrix(){};
   inline SparseMatrix2(Index rows, Index cols):SparseMatrix(rows, cols){};
   inline SparseMatrix2(Index rows, Index cols, StorageIndex* outerIndex):SparseMatrix()
   {
      m_outerSize =IsRowMajor ? rows : cols;
      m_innerSize = IsRowMajor ? cols : rows;
      m_outerIndex = outerIndex;
   };
   inline SparseMatrix2(Index rows, Index cols, StorageIndex* outerIndex, StorageIndex* innerIndex, Scalar* values, int iNNZ):SparseMatrix()
   {
      m_outerSize =IsRowMajor ? rows : cols;
      m_innerSize = IsRowMajor ? cols : rows;
      m_outerIndex = outerIndex;
      ((Storage2*) &m_data)->setExternalData(innerIndex, values, iNNZ);// Acces to protected members of CompressedStorage
   };
   ~SparseMatrix2()
   {
      In destructor we need set NULL pointers for external data. This reasons keep external data from delete operator 
               m_outerSize = 0;
      m_innerSize = 0;
      m_outerIndex = 0;
      ((Storage2*) &m_data)->freeExternalData();// Acces to protected members of CompressedStorage
   };
};
}

This fake wrapper is work in project for future solving System Linear Algebraic Equation like

Code: Select all

int iDOF = solution.Size();
Eigen::SparseMatrix2<double,1,int> A(iDOF, iDOF, ia, ja, a, iNNZ);
Eigen::VectorXd b(iDOF);
for (int ii=0; ii<iDOF; ii++)
{
   b(ii) = rhs[ii];
}
Eigen::ConjugateGradient < CRS, Eigen::Upper, Eigen::DiagonalPreconditioner<double> > solver;
solver.setTolerance(dErr);
solver.setMaxIterations(iIter);
solver.compute(A);
Eigen::VectorXd x = solver.solve(b);


On my opinion using external data in Eigen::SparseMatrix maнbe useful many people.
But using fake wrapper for it is not very good !
I think that in future realeases you create new special template classes for this feature.
Best regards, Majorov S.

You can use Map<SparseMatrix> which is similar in spirit to your wrapper. Then you should still pass SparseMatrix to the solver template parameter:

Code: Select all

Ref<SparseMatrix<double,RowMajor> > mat(rows, cols, nnz, ., ., .);

ConjugateGradient<SparseMatrix<double,RowMajor>, ... > cg(mat);
x = cg.solve(b);


this won't make any copy.

Hi, ggael!
Thank's for answer.
But I understand, that is mean your code

Code: Select all

Ref<SparseMatrix<double,RowMajor> > mat(rows, cols, nnz, ., ., .);


Let I have matrix
[1 0;
0 1];
In code

Code: Select all

int rows = 2;
int cols = 2;
int iNNZ = 2;
int ia[3] = {0, 1, 2};
int ja[2] = {0, 1};
double a[2] ={1,1};


Can you write working code for construct Ref<SparseMatrix<double,RowMajor> >?

oops, I wrote Ref instead of Map (I've edited my previous comment to fix it, then please refer to the doc: http://eigen.tuxfamily.org/dox-devel/cl ... _01_4.html to see the constructor argument order.