[FEA]: Improve thrust::complex compile time throughput

[miscco]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct

Area

Thrust

Is your feature request related to a problem? Please describe.

thrust::complex has an extended API surface compared to cuda::std::complex

The implementation of those additional operators / features has aged a bit and can be improved:

template <typename T0, typename T1>
__host__ __device__ typename detail::disable_if<
  detail::or_<detail::is_same<T0, T1>, detail::not_<detail::is_arithmetic<T1>>>::value,
  complex<typename detail::promoted_numerical_type<T0, T1>::type>>::type
operator+(const T0 &x, const complex<T1> &y)
{
  typedef typename detail::promoted_numerical_type<T0, T1>::type T;
  return complex<T>(x + y.real(), T(y.imag()));
}

This requires us to instantiate is_same and is_arithmetic and promoted_numerical_type even if the first constraint is not true

Describe the solution you'd like

We can avoid extensive instantiations by moving the constraints into the template arguments:

  template <typename T0, typename T1, 
            typename = typename detail::enable_if<detail::not_<detail::is_same<T0, T1>>::value>::type,
            typename = typename detail::enable_if<detail::is_arithmetic<T1>>::value>::type,
            typename = typename detail::enable_if<detail::has_promoted_numerical_type<T0, T1>::value>::type,
>
  __host__ __device__
  complex<typename detail::promoted_numerical_type<T0, T1>::type> &operator=(const U &rhs)
  {
    typedef typename detail::promoted_numerical_type<T0, T1>::type T;
    return complex<T>(x + y.real(), T(y.imag()));
  }

This would short circuit relatively early

Describe alternatives you've considered

No response

Additional context

No response