cuda::std::simd Always use _CCCL_HOST_DEVICE_API

[fbusato]

Description

conservatively use _CCCL_HOST_DEVICE_API (no Tile) for now.

[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: db498d27-eda1-4f34-ba57-dd2437aee42a

📥 Commits

Reviewing files that changed from the base of the PR and between 294326d and c2d9976.

📒 Files selected for processing (2)

• libcudacxx/include/cuda/std/__simd/complex_math.h
• libcudacxx/include/cuda/std/__simd/specializations/fixed_size_mask.h

---

📝 Walkthrough

Summary by CodeRabbit

• Enhancements
  • SIMD complex math operations (real, imaginary, absolute value, argument, norm, conjugate, projection, and transcendental functions) now support execution on both host and device.
  • Mask storage operations enhanced for improved device code compatibility.

Walkthrough

This PR updates API annotations across SIMD complex math operations and mask storage to use _CCCL_HOST_DEVICE_API instead of _CCCL_API. The change applies consistently to unary and binary helper functors, their generator mechanics, all public SIMD vector entry points, and the fixed-size mask accessor, enabling these operations to execute on both host and device without altering function signatures, logic, or return types.

Changes

SIMD Host-Device Callability

Layer / File(s)	Summary
SIMD Complex Unary Operations libcudacxx/include/cuda/std/__simd/complex_math.h	Unary helper functors (__fn_real, __fn_imag, __fn_abs, __fn_arg, __fn_norm, __fn_conj, __fn_proj, __fn_exp, __fn_log, __fn_log10, __fn_sqrt, __fn_sin, __fn_asin, __fn_cos, __fn_acos, __fn_tan, __fn_atan, __fn_sinh, __fn_asinh, __fn_cosh, __fn_acosh, __fn_tanh, __fn_atanh) and internal generator __gen_complex_apply_unary update their operator() annotations. Real-valued vector functions (real, imag, abs, arg, norm) and complex-valued vector functions (conj, proj, exp, log, log10, sqrt, sin, asin, cos, acos, tan, atan, sinh, asinh, cosh, acosh, tanh, atanh) update their declarations. All preserve template constraints, return types, and generator-based construction.
SIMD Complex Binary Operations libcudacxx/include/cuda/std/__simd/complex_math.h	Binary helper functors __fn_polar_binary and __fn_pow_binary, and entry points polar(const basic_vec<_Tp, _Abi>&, const basic_vec<_Tp, _Abi>& = {}) and pow(const basic_vec<_Tp, _Abi>&, const basic_vec<_Tp, _Abi>&) update annotations while maintaining signatures and generator wiring.
Fixed-Size Mask Storage Accessor libcudacxx/include/cuda/std/__simd/specializations/fixed_size_mask.h	The __mask_storage<_Bytes, __fixed_size<_Np>>::__get(__simd_size_type __idx) const noexcept accessor updates its attribute macro.

Possibly Related PRs

• NVIDIA/cccl#8475: Introduced SIMD complex math operations (cuda::std::simd Complex), which are the targets of these annotation updates.

Suggested Reviewers

• wmaxey
• bernhardmgruber
• pciolkosz

---

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

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 1h 21m: Pass: 100%/116 | Total: 1d 09h | Max: 51m 49s | Hits: 97%/348808

See results here.