Re: [eigen] SuperLU

[Gael Guennebaud <gael.guennebaud@xxxxxxxxx>]

Hi,

thanks for reporting this. Indeed, the problem was because I changed
the meaning of Matrix::stride() for vectors... Anyway, problem fixed
now.

cheers,
Gael.


On Thu, Sep 3, 2009 at 5:39 PM, Trevor Irons<trevorirons@xxxxxxxxx> wrote:
> Hi.
>
> Some code I had using the Sparse SuperLU solver quit working with the
> latest update (mercurial tip). I get an error
> "Parameter 13 to routine zgssvx was incorrect"
>
> Based on the SuperLU documentation this is the X vector (SuperMatrix) argument.
> Here is an example piece of code:
>
> #include <fstream>
> #include <Eigen/Core>
> #include <Eigen/Sparse>
> #include <Eigen/Array>
>
> using namespace std;
> using namespace Eigen;
>
> typedef double Real;
> typedef complex<Real> Complex;
>
> int main() {
>
>        cout << "Testing Eigen2 SuperLU\n";
>        int nm = 3*2;
>
>        VectorXcd B(nm);
>        VectorXcd X(nm);
>        SparseMatrix<Complex, RowMajor> A (nm, nm);
>
>        cout << "Building A,B \n";
>        {
>                RandomSetter<SparseMatrix<Complex, RowMajor> > setter(A);
>                for (int i=0; i<nm; ++i) {
>                        B(i) = Complex(1e-7*rand(), 1e-9*rand());
>                        setter(i,i) = Complex(2, .01);
>                        if (i<nm-1)
>                                setter(i,i+1) = Complex(-1, .0);
>                        if (i>0)
>                                setter(i,i-1) = Complex(-1, .0);
>                }
>        }
>        cout << A << endl;
>        cout << B << endl;
>
>        cout << "Solve for X\n";
>        SparseLU<SparseMatrix<Complex>,SuperLU> slu2(A);
>        slu2.solve(B, &X);
>        cout << X << endl;
> }
>
> Compiling with gcc
>
> g++ -I/usr/include/atlas -g -I. -I/usr/include/superlu
> -I/home/tirons/src/eigen2 -DEIGEN_SUPERLU_SUPPORT   -L/usr/lib/
> -lsuperlu /usr/lib/libblas.a  -o test test.cpp -L/usr/lib/
> -lsuperlu /usr/lib/libblas.a
>
> Am I doing something wrong? I get the same error with ColMajor as well.
>
> Thanks,
>
> Trevor
>
>
>