特征将矩阵行/列绑定到向量 l 值参考

Question

How can i pass a column (or row) of a Matrix to a function as an l-value Vector reference? Here is an example for dynamically allocated matrices:

#include "eigen3/Eigen/Eigen"

void f2(Eigen::VectorXd &v) {
  v(0) = 0e0;
}

void f1(Eigen::MatrixXd &m) {
  if (m(0,0) == 0e0) {
    // both calls below fail to compile
    // f2(m.col(0)); // ERROR
    // f2(m(Eigen::all, 0)); //ERROR
  }
  return;
}

int main() {
  Eigen::MatrixXd m = Eigen::MatrixXd::Random(3,3);

  f1(m);
  return 0;
} 

the call to f2 inside f1 triggers a compilation error, of type:

error: cannot bind non-const lvalue reference of type 'Eigen::VectorXd&' {aka 'Eigen::Matrix<double, -1, 1>&'} to an rvalue of type 'Eigen::VectorXd' {aka 'Eigen::Matrix<double, -1, 1>'}

I face the same issue with compile-time sized matrices, eg

constexpr const int N = 3;

void f2(Eigen::Matrix<double,N,1> &v) {
  v(0) = 0e0;
}

void f1(Eigen::Matrix<double,N,N> &m) {
  if (m(0,0) == 0e0) {
    // all calls below fail to compile
    // f2(m.col(0)); ERROR
    // f2(m(Eigen::all, 0));  ERROR
    // f2(m.block<N,1>(0,0)); ERROR
  }
  return;
}

int main() {
  Eigen::Matrix<double,N,N> m = Eigen::Matrix<double,N,N>::Random();

  f1(m);
  return 0;
}  

                                                           

Answer 1

This is what Eigen::Ref is designed to do .

void f2(Eigen::Ref<Eigen::VectorXd> v) {
  v[0] = 123.;
}
void f1(Eigen::Ref<Eigen::MatrixXd> m) {
  m(1, 0) = 245.;
}

int main()
{
    Eigen::MatrixXd m(10, 10);
    f1(m);
    f2(m.col(0));
    assert(m(0,0) == 123.);
    assert(m(1,0) == 245.);

    // also works with parts of a matrix, or segments of a vector
    f1(m.bottomRightCorner(4, 4));
}

Note that mutable references only work if elements along the inner dimension are consecutive. So it works with a single column of a column-major matrix (the default) but not a row. For row-major matrices it is vice-versa.

const refs ( const Eigen::Ref<const Eigen::VectorXd>& ) do work in these cases but they create a temporary copy.