[FEA]: Have a less verbose pinned memory container

[Kh4ster]

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.

Currently, the least verbose way for users to have a container of pinned memory without having to write any custom code is this:

thrust::host_vector<i_t, thrust::mr::stateless_resource_allocator<i_t, thrust::system::cuda::universal_host_pinned_memory_resource>>

See slack discussion here: https://nvidia.slack.com/archives/CCP05T27R/p1723208979166279

Describe the solution you'd like

I think there should be something less verbose like:

thrust::host_vector<i_t, thrust::host_pinned> or thrust::pinned_host_vector<i_t>

Describe alternatives you've considered

Hande written allocator but users shouldn't have to do that.
Libcudf pinned host vector: rapidsai/cudf#16206 rapidsai/cudf#15895 which should be generally available in CCCL.

Additional context

No response

[bernhardmgruber]

There is ongoing work on providing a better replacement for Thrust's vectors that will support pinned memory. @miscco knows more about this.

In the meantime, as a very simple fix, we could just add the following to thrust:

namespace thrust {
  template <typename T>
  using host_pinned_vector = host_vector<T, mr::stateless_resource_allocator<T, system::cuda::universal_host_pinned_memory_resource>>;
}

[bernhardmgruber]

Some pieces were already present in thrust, like universal_host_pinned_memory_resource. The unit tests were also already covering a host pinned vector.

Following existing naming conventions, the new facility will probably be called thrust::universal_host_pinned_vector<T>.