Hi,

I have a piece of code which looks like this:

MatrixXf res;
MatrixXf Y(n2, 3);
for(int i=0; i<n2; i++) {
res = meanValue + k_invQ.at(i).transpose() * correction;
Y(i,0) = res(0,0);
Y(i,1) = res(0,1);
Y(i,2) = res(0,2);
}

where meanValue is a matrix of size (1, 3), k_invQ is a c std vector containing n2 eigen vectors of dynamic size (give by a variable n1) and correction is a matrix of dynamic size (n1, 3). I would like to accelerate this piece of code. For this, I have substituted the above code by:

MatrixXf Y;
VectorXf xx = VectorXf::Ones(n2);
Y = xx * meanValue + k_invQ_Matrix * correction;

where k_invQ_Matrix is a matrix of dynamic size (n2, n1) containing the same information as the one in k_invQ, but in the form of an eigen library matrix. When I run the latter code, the execution time is much larger. Could you please help me with this?

Cheers,
Ricardo

The following should give you the best performance, but make sure you compiled with optimizations enabled (e.g., -O2 with gcc).

Code: Select all

RowVector3f meanValue(...);
MatrixXf Y = meanValue.replicate(n2,1);
Y.noalias() += k_invQ_Matrix * correction;