[eigen] SparseLU generates dense representation somewere

[Wenzel Jakob <wenzel@xxxxxxxxxxxxxxxxxxxx>]

Hello all,

sorry for the potentially very basic question. I'm trying to solve a 
sparse square linear system using Eigen 3.2's new SparseLU method and 
seem to be running into issues. If I compile the snippet below (silly 
but hopefully explains the issue), I get:

$ ./test
terminate called after throwing an instance of 'std::bad_alloc'
  what():  std::bad_alloc
[2]    152688 abort (core dumped)  ./test

So it seems like for some reason it is allocating a dense 
representation. What am I doing wrong?

Thank you,
Wenzel

=========

#define NDEBUG
#include <Eigen/SparseCore>
#include <Eigen/SparseLU>

int main(int argc, char **argv) {
    typedef Eigen::SparseMatrix<double, Eigen::ColMajor> SMatrix;

    const int N = 500000;
    SMatrix A(N,N), B(N, N), C(N, N);

    A.setIdentity();
    B.setIdentity();
    A.makeCompressed();
    B.makeCompressed();

    Eigen::SparseLU<SMatrix> lu;
    lu.compute(A);

    C = lu.solve(B);
}