Hi everybody,

I'm currently debugging my mex file and i'm still facing an issue which is related with the coefficient wise product. As I am not completely comfortable with Eigen, I cannot find the solution. Here is my simplified function (it's a part of a Runge-Kutta algorithm) :

Code: Select all

VectorXd RK4(VectorXd X, struct matrices Par_mat, struct geometrie Par_geo)
{
   int N=Par_geo.Nutiles;
   VectorXd Deplacement(N);
   Deplacement<<X.segment(0,N);
   VectorXd test(N);

   test=Par_mat.Ouv.array()*(1/(Par_geo.heq+Deplacement.array())).array();
   return test;
}

Par_mat.Ouv is a structure variable of type MatrixXd (size (N x 1) ).
Par_geo.heq is a structure variable of type double.
Par_geo.Nutiles is a structure variable of type int.

I believe that the problem is due to the coefficient wise product. I do not understand the reason because both left and right hands are matrix when is use

Code: Select all

m.cwiseProduct(n)

or arrays (in the above example) when i use

Code: Select all

m.array()*n.array()

. There is maybe an issue about the

Code: Select all

double + VectoXd

operation since eigen told me that the type of this sum is 'Const'.

Moreover, the crash is not systematic. Sometimes the code works, and sometimes I have a popup window in Matlab ('Assertion failed ! Program: d:\...\eigen\CwiseBinaryOp.h Line:131 Expression : lhs.rows()==rhs.rows() && lhrs.cols()==rhs.cols()). I guess it is related to a problem of dimension but I checked all my variables and I still do not understand.


Thank you for your help I hope i'm near the end...

yes this means that Ouv and Deplacement do not have the same size. Two suggestions though: (unrelated to your issue)

- Ouv should be a VectorXf
- you can write your expression like this (simpler and better):

Code: Select all

test = Par_mat.Ouv.array() / (Deplacement.array() + Par_geo.heq);


Yes it was a really sneaky problem of size....
It took two hours to find out the solution.

Thank you!