Re: [eigen] Tensor .select() with mixed tensor/scalar arguments

[Rasmus Munk Larsen <rmlarsen@xxxxxxxxxx>]

Yes, select is an operator on tensors. You can write something like

 

ft_result = (ft_A < 0.5f).select(ft_A, ft_A.constant(0.0f));

or you/we could add some syntactic sugar to transform your original _expression_ into this.

On Tue, Nov 5, 2019 at 3:13 AM Wood, Tobias <tobias.wood@xxxxxxxxx> wrote:

Hello,

 

The following simplified program does not compile with Eigen HEAD:

 

#include <unsupported/Eigen/CXX11/Tensor>

 

int main(const int argc, const char *const argv[])

{

  Eigen::Tensor<float, 2> ft_A(2, 4), ft_B(2, 4), ft_result(2, 4);

  ft_A.setRandom();

  ft_B.setZero();

  ft_result = (ft_A < 0.5f).select(ft_A, ft_B);

  ft_result = (ft_A < 0.5f).select(ft_A, 0.f);

}

 

The first call to ft_result compiles fine, but the second does not. The first line of compile errors is:

 

Eigen/src/Core/util/ForwardDeclarations.h:23:47: error: implicit instantiation of undefined template 'Eigen::internal::traits<float>

 

So it looks to me like Eigen tries to treat the 0.f as an Eigen tensor, rather than as the scalar type.

 

Does this count as a bug and if so should I open a bug on the tracker? I looked at trying to fix it myself but sadly this kind of Eigen issue is out of my league.

 

Best wishes,

Toby