[FEATURE]: Expose hash_function member function for hash tables

[PointKernel]

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

The current cuco hash tables don't have hash_function as a member function because we didn't come up with a proper solution for double hashing, i.e., the underlying hash table may utilize more than one hash function so the below interface doesn't seem to be right:

hasher hash_function () const { ... }

Describe the solution you'd like

However, the original idea is not bad and we could use cuda::std::pair<Hash1, Hash2> to define hasher in the case of double hashing:

class static_map {
  ...
  using hasher = cuda::std::conditional_t<is_double_hashing<probing_scheme_type>::value,
    cuda::std::pair<probing_scheme_type::Hash1, probing_scheme_type::Hash2>,
    probing_scheme_type::Hash>;

  ...
  hasher constexpr hash_function () const {
    return cuda::std::pair{probing_scheme.hash_function()};
  }
};

In the future, we can also switch to cuda::std::tuple if more than two hashers are needed.

Describe alternatives you've considered

No response

Additional context

We need a build-time is_double_hashing trait to implement this.

[sleeepyjack]

In the future, we can also switch to cuda::std::tuple if more than two hashers are needed.

I would go with cuda::std::tuple from the start. The only difference is that pair provides .first and .second members. A tuple is more generic and requires no future code change.

[sleeepyjack]

Shall we move this logic to the probing_scheme class? double_hashing::hasher would define tuple<Hash1, Hash2> while the other probing scheme types would just use Hash. From there: auto static_map::hash_function() { probing_scheme_.hash_function(); }

[PointKernel]

Shall we move this logic to the probing_scheme class? double_hashing::hasher would define tuple<Hash1, Hash2> while the other probing scheme types would just use Hash. From there: auto static_map::hash_function() { probing_scheme_.hash_function(); }

Yes, that's what I have in mind as well