Hi, everyone, I have a problem when using UMFPACK in Eigen (with visual studio)

Code: Select all

 Eigen::UmfPackLU<SpMat> solver;
Eigen::SparseMatrix<double> I(num_U,num_U);
I.setIdentity();
A11.makeCompressed();
solver.compute(A11);
A11_inver=solver.solve(I); //A11 is a SPD matrix ,compressed,column-major


It can complie, but get error in the last line of

Code: Select all

   template<typename InputMatrixType>
    void compute(const InputMatrixType& matrix)
    {
      if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
      if(m_numeric)  umfpack_free_numeric(&m_numeric,Scalar());
      grab(matrix.derived());
      analyzePattern_impl();
      factorize_impl();
    }


in UmfPackSupport.h


If I common the code "factorize_impl() " which becomes

Code: Select all

   template<typename InputMatrixType>
    void compute(const InputMatrixType& matrix)
    {
      if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
      if(m_numeric)  umfpack_free_numeric(&m_numeric,Scalar());
      grab(matrix.derived());
      analyzePattern_impl();
        }


no errors , but the solution are all zeros. Maybe there are something wrong with the memory.
Please help me, thanks a lot!

Works for me on Linux and OSX. Can you be more specific about the error you get in factorize_impl?

ggael wrote:Works for me on Linux and OSX. Can you be more specific about the error you get in factorize_impl?

More details, very strange, this code works

Code: Select all

Eigen::UmfPackLU<SpMat> solver
A11.setZero();
for(int i=0;i<A11.cols();i++)
    A11.coeffRef(i,i)=1;
A11.makeCompressed();
solver.compute(A11);
A11_inver=solver.solve(I);


But I replace "A11.setZero();" such as

Code: Select all

Eigen::UmfPackLU<SpMat> solver
for(int i=0;i<A11.cols();i++)
{
   for(int j=0;j<A11.rows();j++)
      A11.coeffRef(i,j)=0;
   A11.coeffRef(i,i)=1;
}
for(int i=0;i<A11.cols();i++)
    A11.coeffRef(i,i)=1;
A11.makeCompressed();
solver.compute(A11);
A11_inver=solver.solve(I);


It can also compile, but get stuck at (in debug)

Code: Select all

inline int umfpack_numeric( const int Ap[], const int Ai[], const double Ax[],
                            void *Symbolic, void **Numeric,
                            const double Control[UMFPACK_CONTROL],double Info [UMFPACK_INFO])
{
  return umfpack_di_numeric(Ap,Ai,Ax,Symbolic,Numeric,Control,Info);
}


It said

When the location "0x0000000000000024" is written to access the conflict


So I really have no idea what happens.

ah, that's probably an issue of umfpack. explicit zeros are pretty bad for sparse solvers.

ggael wrote:ah, that's probably an issue of umfpack. explicit zeros are pretty bad for sparse solvers.

Yes, it should be my mistakes in compile UMFPACK.

I have found an old compiled version of UMFPACK, every thing works well. It must be that I made some mistakes in building UMFPACK.

Again, thank you for your help!