If A is a MatrixXd of size 3x4 and b is a VectorXd of length 4. How can I multiple A and b, element-wise, with b broadcast to every row of A? Something like A.rowwise().cwiseProduct(b) but apparently this doesn't compile.

You are looking for a diagonal product:

A * b.asDiagonal()

An alternative is:

a.array().rowwise() * b.array()

Thanks for the suggestions.

For the first one, would there be performance penalty than simply looping over A's rows (e.g., for (int i = 0; i < A.rows(); ++i) A.row(i).array() *= b.array();)?

For the 2nd one, I got compiling errors: no match for ‘operator*’ in ... Eigen::DenseBase<Derived>::rowwise [...]() * Eigen::MatrixBase<Derived>::array() [...]()’

ggael wrote:You are looking for a diagonal product:

A * b.asDiagonal()

An alternative is:

a.array().rowwise() * b.array()

Diagonal products are fully optimized and vectorized. There is no overhead.

The second version is indeed for the devel version only.