Re: [eigen] Recursion and block matrices

[Norman Goldstein <normvcr@xxxxxxxxx>]

Thanks for the redirect.  Also, Gael's recent example

    VectorXd Block_Solve(Ref<MatrixXd> A, const Ref<const VectorXd>& b ) ...

was very helpful. Here is code that works using Ref:

//////////////////////////////////////////////////////////
#include <Eigen/Core>
using namespace Eigen;

#include <iostream>
using namespace std;

void incr( Ref< MatrixXd >& matV )
{
  ++matV(0,0);

  if( ( 1 == matV.rows() ) ||
      ( 1 == matV.cols() ) )
  {
    return;
  }

  Ref< MatrixXd > sub( matV.block( 1,
                                                         1,
                                                         matV.rows() - 1,
                                                        matV.cols() - 1 ) );
 
  incr( sub );
}// incr


int main( int argc, char* argv[] )
{
  MatrixXd mat( 3, 4 );
  mat << 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, 120;
  cout << "Before:\n" << mat << endl;

  Ref< MatrixXd > sub( mat );
  incr( sub );

  cout << "After:\n" << mat << endl;
  return 0;
}// main
/////////////////////////////////////////////////////////////////////////////////////////////////

I would like to upgrade Ref, further, and have some questions:

-- What was it about the Eigen class designs that made it convenient to restrict Ref to innerStride==1?
Would it be a major development to remove this restriction?  (convenient for transposing and even/odd
subscripting).

-- Is there a signature for incr() that would allow me to pass in the MatrixXd directly, without first
converting it to a Ref in main():
  Ref< MatrixXd > sub( mat );
  incr( sub );

-- Are the two expressions
  Ref< MatrixXd > sub = mat;
  Ref< MatrixXd > sub( mat );
supposed to be equavalent?  I think that a compiler is allowed to interpret the
first one as a default constructor followed by an assignment operator.

Thank you,
Norm


On 09/21/2012 02:09 AM, Christoph Hertzberg wrote:

On 19.09.2012 19:34, Norman Goldstein wrote:

However, there is a software engineering issue that I have not been able
to resolve:

Sorry, if you don't get support re-inventing the wheel ...
If you really have concerns against using the dev-branch (though I can't remember that something bad happened to me using that), you can also just copy the Ref-class from that branch.

Christoph

On 08/28/2012 01:01 AM, Gael Guennebaud wrote:

With the devel branch you might try:

void incr(Ref<MatrixXd> mat) {

// no cast needed
...

}