Hi,

I have the feeling that this is a very easy question, but I couldn't find it in the Tutorial, nor on the board.

I want to compute the
scalar = Vec' * Matrix * Vec;

What I currently do is:
tmp = Vec' * Matrix * Vec;
scalar = tmp(0,0);

But this seems not optimal, and it seems slow.

Thanks, Thomas

Are you using very small fixed size matrix-vector (e.g. Vector4d) or dynamic ones (e.g., VectorXd)? If the former, then you could try to use a lazy product, e.g.:

scalar = (Matrix.lazyProduct(Vec).array() * Vec.array()).sum();

In the former case, the only minor optimization I see is in case you have at hand a vector that you can reuse as a temporary, let's call it aux:
aux.noalias() = Matrix * Vec;
scalar = (Vec.adjoint() * aux);

In both cases, the perf gain will be very small if any.


This is nearly optimal. There is no

Hi ggael,

Thanks for your quick respond.
Good to know it is (near) optimal, it means that another part of my code is (too) slow.

Cheers, Thomas