[libcu++] Improve the implementation of tuple_cat

[miscco]

It relied on recursive types. Rather than that use recursive functions and type aliases.

Every event matching tuple_cat when compiling MatX

tuple_cat were some of the more costly templates in Matx

Inclusive duration, grouped by NVCC primary template / header / function. 506 distinct specializations collapse into these six keys.

Event	Primary key	Baseline (s)	Current (s)	Δ (s)	Count (base → cur)
Instantiating Template Class	__tuple_cat_return_1	195.694	0.000	−195.694	39180 → 0
Instantiating Template Function	__tuple_cat_return_impl	0.000	171.677	+171.677	0 → 38708
Instantiating Template Class	__tuple_cat_return	41.981	0.000	−41.981	7082 → 0
Instantiating Template Class	__tuple_cat_type	3.069	0.000	−3.069	25492 → 0
Processing Header File	tuple_cat.h	3.660	1.530	−2.130	626 → 626
Scanning Function Body	tuple_cat()	0.047	0.046	−0.002	626 → 626

Old recursive class helpers (__tuple_cat_return_1, __tuple_cat_return, __tuple_cat_type) drop to zero. They are replaced by __tuple_cat_return_impl at 171.7 s inclusive — cheaper than the 195.7 s class it replaces, plus the extra ~45 s of the other two helpers.

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

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: 2fca374a-cd0a-4131-bb01-8f6b1f1a636f

📥 Commits

Reviewing files that changed from the base of the PR and between 8da8a83 and c576d42.

📒 Files selected for processing (1)

• libcudacxx/include/cuda/std/__tuple_dir/tuple_cat.h

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

• libcudacxx/include/cuda/std/__tuple_dir/tuple_cat.h

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

---

📝 Summary

Summary by CodeRabbit

• New Features

  • Expanded tuple concatenation support for CUDA tuple-like types, arrays, pairs, complex values, and host standard-library tuples.
  • Added compile-time (constexpr) tuple concatenation support.

• Bug Fixes

  • Improved Microsoft Visual C++ compatibility when working with CUDA and standard arrays.

• Tests

  • Expanded coverage for empty, nested, move-only, reference, ordering, and mixed tuple concatenation scenarios, including host and CUDA tuple-like inputs.

Walkthrough

The change refactors tuple_cat type construction and element assembly, adds MSVC-specific std::array tuple-type support, and expands compile-time and runtime coverage for CUDA and host tuple-like inputs.

Changes

Tuple concatenation

Layer / File(s)	Summary
Tuple type contracts and compiler support libcudacxx/include/cuda/std/__tuple_dir/tuple_cat.h, libcudacxx/include/cuda/std/__tuple_dir/make_tuple_types.h	Return types now derive from merged __tuple_types. CUDA and host std::array tuple-type specializations add an MSVC-compatible path.
Constrained tuple concatenation implementation libcudacxx/include/cuda/std/__tuple_dir/tuple_cat.h	tuple_cat constrains tuple-like inputs and assembles elements through get and forward_as_tuple with forwarding preserved.
Tuple concatenation test coverage libcudacxx/test/libcudacxx/std/utilities/tuple/tuple.tuple/tuple.creation/tuple_cat.pass.cpp	Tests now run at compile time and runtime and cover CUDA, host, empty, nested, reference, movable, array, complex, pair, and tuple-like inputs.

Suggested reviewers: davebayer

Priority: ⬇️ Low

Change: Refactor

Merge Risk: 🔵 Low · up to c576d

The remaining concern is limited to annotated preprocessor comment consistency in a tuple type-support header and does not indicate a functional regression.

---

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

[coderabbitai]

🧹 Nitpick comments (1)

libcudacxx/include/cuda/std/__tuple_dir/make_tuple_types.h (1)

72-77: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

suggestion: Use the exact #if _CCCL_COMPILER(MSVC) condition in each annotated #else and #endif comment. Do not use the negated !_CCCL_COMPILER(MSVC) form.

• libcudacxx/include/cuda/std/__tuple_dir/make_tuple_types.h#L72-L77: update the #else and #endif comments.
• libcudacxx/include/cuda/std/__tuple_dir/make_tuple_types.h#L92-L97: update the #else and #endif comments.

Source: Learnings

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: a3326b36-2078-4179-b9f7-7309e33c6dec

📥 Commits

Reviewing files that changed from the base of the PR and between d8a7407 and 8da8a83.

📒 Files selected for processing (1)

• libcudacxx/include/cuda/std/__tuple_dir/make_tuple_types.h

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 3h 07m: Pass: 100%/195 | Total: 2d 23h | Max: 3h 06m | Hits: 74%/792259

See results here.