[FEA]: optimize the case of storing a `resource_ref` in an `any_resource`

[ericniebler]

Is this a duplicate?

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

Area

libcu++

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

with the current implementation, storing a cuda::rm::resource_ref in a cuda::experimental::any_resource will double- type erase the underlying type, leading to double indirections for every function call.

Describe the solution you'd like

find a way to bring this scenario down to a single indirection per type-erased function call.

Describe alternatives you've considered

No response

Additional context

No response

[miscco]

Yeah I was thinking about that too. I fear that the "correct" solution is to add a second non-type template argument that signals whether its a reference or owning and then conditionally do the right thing in the constructor, etc..

That would allow us to use a clean base class and implement all the interdependencies in one place