When using the Eigen wrapper for SuperLU with optimizations enabled in Visual C++ 2005 SP1, the result vector doesn't contain the solution to the problem. The initial value of the x vector determines what it contains after the the solve function is finished. I've used SuperLU before in other projects (but without Eigen) and never had this problem.

Both succeeded() and solve() returns true.

When using:
x = Eigen::VectorXf::Constant(b.rows(), 5.0f);

I get:
...
23.3701
80.9484
56.7775
35.2517
18.4563
100.196
12.018
119.995

When using
x = DenseVector(b.rows());
x.setZero();

I get:
...
-0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

This only happens when /O1,/O2 or /Ox are enabled. When these are not enabled (/Od) I get the correct result:
...
0.0168678
0.0747107
0.0517906
0.0197386
0.0289349
0.00835007
0.0805177
0.0658547
0.0773363
0.0338781
0.0553986
0.034374
0.0601

Edit: It works when I use x = b, but why it this necessary?

I added the following code to the solver routine ( in sgssvx.c ):
if( Bstore->nzval && Xstore->nzval )
{
assert( Bstore->nzval != Xstore->nzval );
}

The assert is only triggered when using any of the optimization modes.

The problem: SluMatrix needs an operator=