Hi everybody,

I am using Eigen 3.0.3 and until now it has worked pretty well.

I have a strange problem with a multiplication between two matrices. The two matrices are initialized just before the multiplication. The result of the multiplication is wrong for a part. The beginning is okay but not the end and I don't know why. If I print these two matrices just before the multiplication they are okay but the result is not.

Here is the code not working well:

Code: Select all

int sizeStateVector = 6+3*state_.getNbLandmarks();
// typedef Matrix<double, 2, Dynamic> Matrix2Nd;
Matrix2Nd H(2, sizeStateVector);
H.setZero();
jacobianH(i, H);
MatrixXd P(sizeStateVector, sizeStateVector);
state_.cov(P);

// okay
cout << P << endl << endl;
cout << H.transpose() << endl << endl;
// not okay
cout << P*H.transpose() << endl;


I know there are some restrictions on how to use matrices as function parameters, and I have tried different ways, but as the matrices seem right, I don't see where the error could be.

Maybe I am missing something but if you could help me with that, It'd be nice.

well, maybe you could try to run your program under valgrind to detect possible memory errors.

Sorry, I forgot to mention that I already checked with valgrind and have no memory leaks.

Thanks for the help anyway.

Some tests:

What is the typical value of sizeStateVector ?

Do you get the same results with a MatrixXd instead of a Matrix2Nd ?

What if you disable vectorization by compiling with -DEIGEN_DONT_VECTORIZE ?

What if you try with P.lazyProduct(H.transpose()) ?

Typical value for sizeStateVector is between 6 and 12 here.

Yes, same results with MatrixXd.

Still the same results with -DEIGEN_DONT_VECTORIZE.

Once again, still the same results with the lazy product.

Not good, huh? Thank you for your time.