Certain operations involving SitePropagator::scalar_object won't compile with CUDA for Nc > 3

[edbennett]

In trying to understand aportelli/Hadrons#94, it transpires that certain arithmetic operations in Grid will not compile for SU(4).

For example, the following test program gives two build errors, one caused by line 6 and one by line 15:

#include <Grid/Grid.h>

using namespace Grid;

void test(WilsonImplR::PropagatorField &out, WilsonImplR::ComplexField in, const WilsonImplR::SitePropagator::scalar_object idMat) {
  out = in * idMat;
}

void test2(WilsonImplR::SitePropagator::scalar_object sp1,
	   WilsonImplR::PropagatorField pf1,
	   WilsonImplR::PropagatorField pf2,
	   Gamma g1,
	   Gamma g2)
{
  LatticeComplex out = trace(g1 * sp1 * adj(pf1) * (g1 * g2) * pf2);
}

In both cases, using an auto declaration works, but then assigning back to the desired type re-triggers the error, so the cast appears to be at least part of the problem.

I compiled with:

nvcc -x cu -I${HOME}/prefix_su4test/include -ccbin mpicxx -I${HOME}/prefix/include --expt-extended-lambda -c -o test.o test.cpp

These examples compile fine for SU(3), and also work fine with SU(4) compiled for CPU.

Is this something that it would be possible to fix, please?