Add a missing RMM_EXEC_CHECK_DISABLE in device_uvector constructors

[caugonnet]

Description

One of the constructor was missing a RMM_EXEC_CHECK_DISABLE annotation, resulting in (false positive ?) warnings:

error: calling a __host__ function("rmm::device_buffer::operator =(rmm::device_buffer&&)") from a __host__ __device__ function("rmm::device_uvector<unsigned long> ::operator =") is not allowed

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[bdice]

Do we need the same change in device_scalar? The pattern looks identical.

Can you provide a reproducer of the error, perhaps as a test that can be compiled?

[harrism]

Yeah what is the false positive? Is relaxed-contstexpr-defined?

[caugonnet]

I could not make a repro yet, and the actual offending code is cuGraph, so nothing close to a simple example to reproduce (it does use relaxed constexpr indeed).

Doing some archeology, the reason for disabling these warning in the other constructor was the following (even if i could not make a repro with thrust::optional<...> yet.

commit dc3ca447615518f0cdfcc45127eb63868d07ce63
Author: Robert Maynard <[email protected]>
Date:   Thu Mar 18 09:09:15 2021 -0700

    device_uvector can be used within thrust::optional (#734)
    
    Fixes https://github.com/rapidsai/cudf/issues/7629
    
    Since thrust::optional is marked with host and device markup,
    we need to make sure we don't generate errors when device_uvector
    is placed inside of it as it will inherit the device markup.