libcu++: atomic backend naming unification.

[griwes]

Description

This is a mechanical extraction of the renaming aspects of #10908 into its own PR.

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[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: 15d97478-f154-4f7a-85ae-5cdf557d05c0

📥 Commits

Reviewing files that changed from the base of the PR and between 452c02a and febfb73.

📒 Files selected for processing (13)

• libcudacxx/codegen/generators/compare_and_swap.h
• libcudacxx/codegen/generators/definitions.h
• libcudacxx/codegen/generators/exchange.h
• libcudacxx/codegen/generators/fence.h
• libcudacxx/codegen/generators/fetch_ops.h
• libcudacxx/codegen/generators/ld_st.h
• libcudacxx/include/cuda/__atomic/atomic.h
• libcudacxx/include/cuda/std/__atomic/functions/cuda_local.h
• libcudacxx/include/cuda/std/__atomic/functions/cuda_ptx_derived.h
• libcudacxx/include/cuda/std/__atomic/functions/cuda_ptx_generated.h
• libcudacxx/include/cuda/std/__atomic/functions/cuda_ptx_generated_helper.h
• libcudacxx/include/cuda/std/__atomic/order.h
• libcudacxx/include/cuda/std/__atomic/types/base.h

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

---

📝 Walkthrough

Summary by CodeRabbit

• Refactor
  • Standardized CUDA atomic operation naming across fences, loads, stores, exchanges, compare-exchanges, and fetch operations.
  • Updated memory-order and operand terminology for consistency.
  • Preserved existing atomic behavior, including volatile access, weak-operation handling, and host-side dispatch.
  • Updated device-side operations to use the standardized CUDA atomic interfaces.

Walkthrough

Changes

The PR renames generated CUDA atomic helpers, internal order and operand entities, local atomic helpers, and device dispatch calls from legacy CUDA prefixes to the __cuda_atomic_* naming scheme. Operation behavior and host-side dispatch remain unchanged.

CUDA atomic naming alignment

Layer / File(s)	Summary
Atomic naming contracts libcudacxx/codegen/generators/definitions.h, libcudacxx/include/cuda/std/__atomic/functions/cuda_ptx_generated_helper.h, libcudacxx/include/cuda/std/__atomic/order.h	Internal order, MMIO, operand, deduction, type-selection, and stronger-order entities use the __cuda_atomic_* names.
Generated dispatch updates libcudacxx/codegen/generators/{compare_and_swap,exchange,fence,fetch_ops,ld_st}.h	Generated load, store, exchange, compare-exchange, fetch, and fence helpers use the renamed dispatch, order, operand, proxy, and local-fast-path identifiers.
Device atomic helper updates libcudacxx/include/cuda/std/__atomic/functions/cuda_local.h, libcudacxx/include/cuda/std/__atomic/functions/cuda_ptx_derived.h	CUDA-local operations and PTX-derived fetch-update paths call the renamed helpers and retain existing operation behavior.
Public device dispatch wiring libcudacxx/include/cuda/__atomic/atomic.h, libcudacxx/include/cuda/std/__atomic/types/base.h	Device fences, compare-exchange operations, and fetch operations invoke the renamed CUDA dispatch functions. Host dispatch remains unchanged.

Suggested reviewers: fbusato, miscco

Merge Risk: ⚪ Minimal · up to febfb

This PR performs a localized atomic backend naming update with no actionable merge-blocking risk remaining beyond normal checks and review.

---

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

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 2h 24m: Pass: 100%/117 | Total: 3d 03h | Max: 1h 30m | Hits: 59%/646965

See results here.

[davebayer]

I'm not a fan of using __cuda prefix, I'd much rather see something like __cccl_atomic

[griwes]

My logic here is that this means "this is a part of the cuda::atomic implementation". But additionally, __cuda_atomic just codifies an existing naming scheme that existed at the top layer of the backend; if we move to __cccl, this will be a larger rename.

[miscco]

I am a bit torn here.

We usually want to prefix with __cccl rather anything else to make sure where something belongs.

However, this is the implementation of cuda::atomic so this makes it appropriate in this context

[griwes]

@miscco @wmaxey just wanted at least one of you to take a look, since we touched on this offline.

[miscco]

Fine with the naming.