Add memory resource parameter to cudf::hash_join constructors

[PointKernel]

Description

Follow-up to #23220.

This PR adds an mr parameter to the cudf::hash_join constructors to control the memory usage of the internal cuco data structures, to better align with the temporary mr effort (#20780). The resource must remain valid for the lifetime of the hash_join object.

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]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[PointKernel]

/ok to test 8e12597

[PointKernel]

copydoc causes Doxygen check failures, and the only way to fix them is to duplicate the documentation instead of using copydoc.

[PointKernel]

/ok to test c2c5d7d

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 8a10a59d-7665-4efb-87b4-65502396481e

📥 Commits

Reviewing files that changed from the base of the PR and between c2c5d7d and 269cfbb.

📒 Files selected for processing (3)

• cpp/include/cudf/detail/join/hash_join.hpp
• cpp/include/cudf/join/hash_join.hpp
• cpp/src/join/hash_join/hash_join.cu

🚧 Files skipped from review as they are similar to previous changes (1)

• cpp/include/cudf/join/hash_join.hpp

---

📝 Walkthrough

Summary by CodeRabbit

• New Features
  • Hash joins now accept an optional device memory resource parameter to control temporary GPU memory allocations.
  • Default behavior remains unchanged when the memory resource is not specified.
• Tests
  • Added coverage to verify memory-resource statistics are updated during hash-join construction and execution.
• Documentation
  • Updated the public API documentation for hash-join constructors to include the new memory-resource parameter.

Walkthrough

Hash join constructors now accept an explicit device memory resource, propagate it to internal hash-table allocations, and retain default-resource behavior for public overloads. A test verifies allocation tracking and inner-join results with a statistics resource adaptor.

Changes

Hash Join Memory Resource

Layer / File(s)	Summary
Public constructor contract cpp/include/cudf/detail/join/hash_join.hpp, cpp/include/cudf/join/hash_join.hpp	Public and detail constructors document and accept the additional device memory resource parameter, with a current-device-resource default on public overloads.
Internal allocation propagation cpp/src/join/hash_join/hash_join.cu	Constructor definitions and declarations propagate the resource into the internal hash join and use it for the Cuco hash-table allocator.
Memory resource validation cpp/tests/join/join_tests.cpp	A test uses a statistics resource adaptor, checks peak allocation tracking, runs an inner join, and validates the output.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Suggested reviewers: wence-, misiugodfrey, mhaseeb123

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: adding a memory resource parameter to cudf::hash_join constructors.
Description check	✅ Passed	The description is directly related to the changeset and matches the implemented memory resource update and tests.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@cpp/include/cudf/detail/join/hash_join.hpp`:
- Around line 71-82: Replace the `@copydoc` directive above the detail-header
hash_join overload with duplicated documentation matching the corresponding
public declaration, including its parameter descriptions and the existing
load_factor documentation. Update only the documentation for hash_join; retain
the declaration and signatures unchanged.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 09118cc4-427a-46fb-974e-790095f0c448

📥 Commits

Reviewing files that changed from the base of the PR and between 151de69 and c2c5d7d.

📒 Files selected for processing (4)

• cpp/include/cudf/detail/join/hash_join.hpp
• cpp/include/cudf/join/hash_join.hpp
• cpp/src/join/hash_join/hash_join.cu
• cpp/tests/join/join_tests.cpp

[bdice]

This seems almost-right. We're using the passed-in mr because this is object-oriented, so we're treating the hash table more like output memory than temporary memory. However, this also needs to store a cuda::mr::any_resource<device_accessible> as a member of the cudf::hash_join object -- we don't want to put requirements on the lifetime of a ref. The resource should be reified and kept alive by the object holding the memory.

Compare to rmm::device_buffer, because this class is now more like a data container. https://github.com/rapidsai/rmm/blob/main/cpp/include/rmm/device_buffer.hpp

For container-like classes, we use cuda::mr::any_resource<cuda::mr::device_accessible> as the mr parameter rather than a ref, to allow construction from resource rvalues to avoid making a copy. Then store _mr{std::move(mr)} as a member.

[PointKernel]

@bdice Good point on ownership. Rather than adding a member, I switched the constructor mr parameter to cuda::mr::any_resource<cuda::mr::device_accessible> (matching rmm::device_buffer/device_uvector) and move it into the cuco map's allocator. rmm::mr::polymorphic_allocator already stores an owning any_resource member, and the map holds the allocator by value, so the resource is reified and kept alive for exactly the hash table's lifetime without a separate hash_join member.

[bdice]

Excellent, thanks for the ownership fix!

[PointKernel]

/merge