Hey

I need to do a matrix-vector multiply, using the matrix or it's transpose depending on some condition, and I wanted to hide this test so I came up with: (where my_c is an instance of class c containing some matrices, and v is a vector)

Code: Select all

x = my_c.get_m()*v


the implementation of c::get_m():

Code: Select all

const MatrixXf & c::get_m(){
  if (cond)
    return m;
  else
    return m.transpose();
}


This works OK. My first question is if this code will be good and do some kind of implicit transposition. The next thing is, I also want to sometimes modify either the matrix or its transpose depending on a condition i.e. do something like

Code: Select all

my_c.get_m() += /* matrix expression */


But I found that when I alter get_m to instead return a regular, non-const, reference I get an error

Code: Select all

error: initial value of reference to non-const must be an lvalue


Is there some way I can return a non-constant reference to the transpose?

Thanks in advance for any insights

MatrixXf is a column major matrix, so when you do return m.transpose();, m.transpose() will be copied into a temporary. If "cond" can be known at compile time, then you can use template meta programming to return a const MatrixXf& or a Transpose<const MatrixXf> depending on "cond".

OK Thanks for the help