Code: Select all

template <typename Derived1,typename Derived2>
typename Derived1::Array my_saxpy(const double& a,const ArrayBase<Derived1>& x,const ArrayBase<Derived2>& y)
{
  ArrayXXd a_array;
  a_array = ArrayXXd::Constant(x.rows(),x.cols(),a);
  return a_array/x + y;
}

int main()
{
  int N(800);
  double a(2.);
  ArrayXXcd x(N,N),y(N,N);
  x.setOnes();
  y.setOnes();
  std::cout << "my_saxpy(a,x,y)(0,0) = " << (my_saxpy(a,x,y))(0,0) << std::endl;
 
}


The above code snippet shows a simple function to calculate (a/x+y) for the ArrayXXcd class. The program fails to compile with the error:

error: static assertion failed with "YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY"
EIGEN_CHECK_BINARY_COMPATIBILIY(BinaryOp,typename Lhs::Scalar,typename Rhs::Scalar);

If instead, I try and compile the function with the usual SAXPY function == a*x+y, the code works perfectly fine. It appears that the division operator is not properly overloaded when complex types are mixed with reals. I have also tried the cast() method using .cast<std::complex<double>>() but I get a 'type does not exist error'. Can you suggest a work around?

Thanks

What about:

a * x.inverse() + y

Thanks gael!

Your suggestion works. However, I tried playing around with the expression. I run into the same problem when I try and calculate:

Code: Select all

a_array + x + y;


What do I do here? Is there a general way to make complex types play nice with real ones?

yes, it's currently limited to products. Again, no need to use a_array, complex<double>(a) + x + y should work. For an expression, a_array.cast<complex<double> >() will do the conversion.

Much appreciated!