Let's say you have an (m x n) matrix M and a vector V of length n, and you want to multiply each row of M by the values in V, elementwise. With broadcasting, you would write:

Code: Select all

MatrixXf M(m, n);
VectorXf V(n);

...

M.rowwise() *= V;


Mathematically, you could also treat V as a square, diagonal matrix and right-multiply it, and this should achieve the same thing:

Code: Select all

M *= V.asDiagonal();


Is there a functional difference between these two? Do they generate the same code?

Cheers
Leo

Code: Select all

M *= V.asDiagonal();


means

Code: Select all

M = M * V.asDiagonal();


which is functionally equivalent to

Code: Select all

M.colwise() *= V;


If you want rowwise, you have to put the diagonal on the left:

Code: Select all

M = V.asDiagonal() * M;