Enable __float128 in cuda::std::copysign and cuda::std::signbit

[temujinkz]

Adds __float128 support to cuda::std::copysign and test coverage for both copysign and signbit. Doing these two together since the signbit test generates its negative values through copysign.

copysign follows the same pattern as fabs in #9895: route to __builtin_copysignf128 when available, to __nv_fp128_copysign in device code otherwise, and keep the existing storage mask path as the constexpr fallback. signbit needed no implementation change, the generic sign mask path already handles binary128.

No __float128 capable host here (macOS arm64), so locally I could only compile and run both tests with float128 disabled to check for regressions. CI is the gate for the new paths.

Fixes #9961, fixes #9967

[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.

[coderabbitai]

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes

  • Improved copysign handling for __float128 values across supported host and device environments.
  • Enhanced sign-bit processing for extended-precision floating-point values.

• Tests

  • Added coverage for copysign and signbit operations using __float128 when available.

Walkthrough

Changes

Float128 math support

Layer / File(s)	Summary
__float128 copysign dispatch libcudacxx/include/cuda/std/__cmath/copysign.h	Adds guarded copysignf128 builtin and NV device-intrinsic paths for __float128, while retaining generic floating-point handling for other types.
Float128 validation libcudacxx/test/libcudacxx/std/numerics/c.math/fp_manip/copysign.pass.cpp, libcudacxx/test/libcudacxx/std/numerics/c.math/fp_traits/signbit.pass.cpp	Adds guarded __float128 coverage to the copysign and signbit tests.

Assessment against linked issues

Objective	Addressed	Explanation
Enable __float128 in cuda::std::copysign [#9961]	✅
Enable __float128 in cuda::std::signbit [#9967]	❓	Only test coverage changes are shown; no signbit implementation change is provided, so existing implementation support cannot be confirmed.

Suggested reviewers: gevtushenko

---

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

[coderabbitai]

🧹 Nitpick comments (1)

libcudacxx/include/cuda/std/__cmath/copysign.h (1)

77-95: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

suggestion: Keep the preprocessor annotations identical to their controlling condition.

Line 80 uses #if defined(_CCCL_BUILTIN_COPYSIGNF128), but Lines 88 and 93 annotate a negated form. Repeat defined(_CCCL_BUILTIN_COPYSIGNF128) verbatim in both comments.

Based on learnings, annotated #else and #endif branches must repeat the exact condition text from the corresponding #if.

Source: Learnings

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 317e6c30-970e-4f59-b80c-0f98030a808d

📥 Commits

Reviewing files that changed from the base of the PR and between a129c6d and b7dbc5e.

📒 Files selected for processing (3)

• libcudacxx/include/cuda/std/__cmath/copysign.h
• libcudacxx/test/libcudacxx/std/numerics/c.math/fp_manip/copysign.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/c.math/fp_traits/signbit.pass.cpp