[Tile] Actually rewrite complex for extended floating point types once more

[miscco]

This ensures that it works fine, although it has a minor perf regression

[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: 6da3f4b6-b68c-4d73-871c-0a381932fc2e

📥 Commits

Reviewing files that changed from the base of the PR and between 8548977 and b8b8c7c.

📒 Files selected for processing (80)

• libcudacxx/include/cuda/std/__cccl/extended_data_types.h
• libcudacxx/include/cuda/std/__complex/complex.h
• libcudacxx/include/cuda/std/__complex/nvbf16.h
• libcudacxx/include/cuda/std/__complex/nvfp16.h
• libcudacxx/test/libcudacxx/std/numerics/bit/bit.cast/bit_cast.trivially_copyable.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cases.h
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/arg.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/conj.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/imag.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/norm.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/pow.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/proj.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/real.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/assignment_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/assignment_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/divide_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/divide_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/minus_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/minus_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/plus_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/plus_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/times_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/times_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.members/construct.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.members/real_imag.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_divide_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_divide_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_equals_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_minus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_minus_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_not_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_not_equals_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_plus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_plus_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_times_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_times_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_divide_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_minus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_not_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_plus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_times_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/stream_input.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/stream_output.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/unary_minus.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/unary_plus.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/acos.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/acosh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/asin.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/asinh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/atan.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/atanh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/cos.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/cosh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/exp.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/log.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/log10.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/pow_complex_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/pow_complex_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/pow_scalar_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/sin.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/sinh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/sqrt.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/tan.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/tanh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.tuple/get.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.tuple/tuple_element_compiles.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.tuple/tuple_size_compiles.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/abs.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/arg.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/conj.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/imag.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/norm.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/polar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/real.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex/abi_latest.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex/traits.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex/types.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/layout.pass.cpp

🚧 Files skipped from review as they are similar to previous changes (75)

• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_times_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_minus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/unary_minus.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/norm.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_not_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/real.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_divide_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/minus_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/layout.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/atanh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/cos.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/imag.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/divide_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_divide_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_minus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/tan.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/asin.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/assignment_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_minus_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex/traits.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/pow_complex_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/minus_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/acosh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_equals_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.members/construct.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/pow.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_plus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/polar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/acos.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.members/real_imag.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/log.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.tuple/tuple_element_compiles.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/pow_complex_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/times_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cases.h
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/exp.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/times_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_plus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/sqrt.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_not_equals_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/log10.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/sin.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/arg.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/plus_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/tanh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_times_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_not_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/proj.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/sinh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_divide_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/plus_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/arg.pass.cpp
• libcudacxx/include/cuda/std/__cccl/extended_data_types.h
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.tuple/get.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/unary_plus.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex/abi_latest.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/norm.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/abs.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/imag.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.tuple/tuple_size_compiles.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/divide_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex/types.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_plus_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/real.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/conj.pass.cpp
• libcudacxx/include/cuda/std/__complex/complex.h
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/cosh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_times_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/pow_scalar_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/atan.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/conj.pass.cpp
• libcudacxx/include/cuda/std/__complex/nvfp16.h
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/asinh.pass.cpp

---

📝 Walkthrough

Summary by CodeRabbit

• New Features
  • Improved complex-number support for half-precision and bfloat16 values during tile compilation.
  • Added broader cross-type construction and assignment support for extended floating-point complex values.
  • Improved tile-compatible arithmetic and component-wise operations for half and bfloat16 complex numbers.
• Bug Fixes
  • Corrected scalar complex division assignment availability across compilation targets.
• Tests
  • Expanded complex arithmetic, transcendental, conversion, tuple, layout, and bit-casting coverage for tile compilation and supported extended floating-point types.

Walkthrough

Extended floating-point detection and cuda::std::complex half and bfloat16 support now include tile compilation. Tile-specific representations and component-wise operations support these builds. Tests for bit casts, types, operators, values, and transcendental functions now run when the feature macros are enabled.

Changes

Extended complex tile support

Layer / File(s)	Summary
Complex representations and APIs libcudacxx/include/cuda/std/__cccl/extended_data_types.h, libcudacxx/include/cuda/std/__complex/complex.h, libcudacxx/include/cuda/std/__complex/nvbf16.h, libcudacxx/include/cuda/std/__complex/nvfp16.h	Float128 detection no longer directly excludes tile compilation. Half and bfloat16 complex types use tile-specific representations and component-wise operations. Cross-type constraints and API annotations were updated.
Type and bit-cast coverage libcudacxx/test/libcudacxx/std/numerics/bit/bit.cast/bit_cast.trivially_copyable.pass.cpp, libcudacxx/test/libcudacxx/std/numerics/complex.number/cases.h, libcudacxx/test/libcudacxx/std/numerics/complex.number/complex/*.pass.cpp, libcudacxx/test/libcudacxx/std/numerics/complex.number/layout.pass.cpp	Bit-cast, comparison, layout, trait, ABI, and type tests no longer exclude tile compilation.
Complex operators and accessors libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/*, libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/*, libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.members/*, libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/*, libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.tuple/*	Half and bfloat16 tests for construction, assignment, arithmetic, comparisons, streams, accessors, and tuple operations now run during tile compilation.
Complex math coverage libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/*, libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/*	Half and bfloat16 value and transcendental tests, including edge cases and expected-value calculations, no longer exclude tile compilation.

Suggested labels: libcu++

Suggested reviewers: pciolkosz, gevtushenko

---

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

[coderabbitai]

Actionable comments posted: 2

🧹 Nitpick comments (1)

libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/plus_equal_scalar.pass.cpp (1)

45-50: 🎯 Functional Correctness | 🔵 Trivial

suggestion: Run the affected tile tests. Use the relevant CCCL CMake preset and targeted test targets to compile and execute the new __half and __nv_bfloat16 instantiations. The supplied context does not include build or test results. As per coding guidelines, validate changes with targeted builds and tests using the provided CMake presets. As per path instructions, use .agent/skills/cccl-test/SKILL.md and its routed references for libcudacxx complex-number tests.

Sources: Coding guidelines, Path instructions

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: cf1d5208-a708-426c-864b-2362370af637

📥 Commits

Reviewing files that changed from the base of the PR and between 0c7c3ec and 5079a40.

📒 Files selected for processing (80)

• libcudacxx/include/cuda/std/__cccl/extended_data_types.h
• libcudacxx/include/cuda/std/__complex/complex.h
• libcudacxx/include/cuda/std/__complex/nvbf16.h
• libcudacxx/include/cuda/std/__complex/nvfp16.h
• libcudacxx/test/libcudacxx/std/numerics/bit/bit.cast/bit_cast.trivially_copyable.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cases.h
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/arg.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/conj.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/imag.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/norm.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/pow.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/proj.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/cmplx.over/real.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/assignment_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/assignment_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/divide_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/divide_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/minus_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/minus_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/plus_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/plus_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/times_equal_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.member.ops/times_equal_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.members/construct.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.members/real_imag.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_divide_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_divide_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_equals_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_minus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_minus_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_not_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_not_equals_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_plus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_plus_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_times_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/complex_times_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_divide_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_minus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_not_equals_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_plus_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/scalar_times_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/stream_input.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/stream_output.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/unary_minus.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.ops/unary_plus.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/acos.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/acosh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/asin.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/asinh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/atan.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/atanh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/cos.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/cosh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/exp.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/log.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/log10.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/pow_complex_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/pow_complex_scalar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/pow_scalar_complex.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/sin.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/sinh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/sqrt.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/tan.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.transcendentals/tanh.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.tuple/get.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.tuple/tuple_element_compiles.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.tuple/tuple_size_compiles.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/abs.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/arg.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/conj.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/imag.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/norm.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/polar.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex.value.ops/real.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex/abi_latest.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex/traits.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/complex/types.pass.cpp
• libcudacxx/test/libcudacxx/std/numerics/complex.number/layout.pass.cpp

[griwes]

The question below, and the direct code suggestion, should be resolved before merging. The other suggestion is up to you.

Tests look good, except apparently you've not updated complex.value.opts/proj.pass.cpp; divide_equal_complex.pass.cpp also has an issue, but that one got cleanly caught by CI.

[griwes]

Suggestion: I started five separate comments recommending to un-special-case this and all of them failed. I hate it here. Best I can think of is having a separate template<typename T> struct __type_to_storage_vector { using __type = typename __type_to_vector<T>::__type; };, in these tile paths specializing for half and bf16, and then using __type_to_storage_vector throughout instead of the special cases here and in half? But honestly I am not sure if that is at all better. Sigh. Though it does mean we're not typedefing the same name to different types in the different branches.

[miscco]

I mean there are non different branches, there is either tile mode or not

[griwes]

What I mean is that there's two separate definitions of __complex_nv_bfloat_repr_t, and we could potentially fold it, but this is very much not a blocker.

[coderabbitai]

Note

GitHub couldn't provide a complete incremental comparison for this pull request, so CodeRabbit is performing a full review instead. This review may take a little longer.

[github-actions]

⏱️ CCCL compile-time benchmark comparison: Public headers compile-time bench

Result: 0 regression row(s), 6 improvement row(s) above threshold.

Run	Value
Config	public-headers-gcc13
Baseline	origin/main
Preset	all-dev
Targets	cub.headers.base, thrust.cpp.cuda.headers.base, libcudacxx.test.public_headers
GPU / launch args	rtx2080 / --cuda 13.3 --host gcc13

Artifacts: reports and traces

Direct file processing

-f file-processing exclusive --sort total

🟢 Direct file processing — Improvements

Rank	Improvement impact	Selected Δ	Baseline	Current	Event	Matched traces
1	0.772741	-0.772741	5.267038	4.494297	Processing Header File: libcudacxx/include/cuda/std/__cccl/prologue.h	550
2	0.406431	-0.406431	6.648210	6.241779	Processing Header File: libcudacxx/include/cuda/std/__type_traits/type_list.h	477
3	0.354418	-0.354418	6.285254	5.930836	Processing Header File: libcudacxx/include/cuda/std/__iterator/concepts.h	437
4	0.305156	-0.305156	4.231467	3.926311	Processing Header File: cub/cub/util_type.cuh	279
5	0.302361	-0.302361	1.872687	1.570326	Processing Header File: libcudacxx/include/cuda/std/__cccl/epilogue.h	550
6	0.255213	-0.255213	10.535898	10.280685	Processing Header File: libcudacxx/include/cuda/__device/physical_device.h	94

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 4h 39m: Pass: 100%/115 | Total: 4d 20h | Max: 4h 38m | Hits: 45%/1280044

See results here.

AI failure analysis

1. RAPIDS environment merges incompatible rapids-logger pins · 2 jobs

Explanation: Both RAPIDS matrices fail before compilation because the generated Conda environment simultaneously requires incompatible `rapids-logger` 0.2 and 0.3 series. The cugraph/wholegraph matrix additionally exposes CUDA 13.3 compatibility constraints while resolving cuML, but its viable candidates still require `rapids-logger` 0.2 and are blocked by the same 0.3 pin.

Evidence:

Build RAPIDS (optional) / rmm ucxx raft cuvs cugraph wholegraph, step 6

2026-08-11T09:19:05.7100753Z error    libmamba Could not solve for environment specs
2026-08-11T09:19:05.7107689Z     ├─ cuda-version =13.3 * is requested and can be installed;
2026-08-11T09:19:05.7126993Z     │  └─ cuml [26.10.00a52|26.10.00a54|26.10.00a55] would require

Root cause: The selected RAPIDS 26.10 development branches contribute dependency files pinned to different major-minor `rapids-logger` series, and the build utility merges them into one unsatisfiable environment. The logs do not attribute each pin to its originating cloned repository, so the exact upstream dependency file must be identified during reproduction; the PR diff only changes libcudacxx code and does not modify this CI configuration. Sources: .github/workflows/build-rapids.yml:67, .github/workflows/build-rapids.yml:68, ci/rapids/post-create-command.sh:38.

Suggested next steps: Regenerate each failing matrix environment while adding one RAPIDS repository at a time to identify which dependency files request `rapids-logger==0.2.*` and `==0.3.*`, then align those repositories on one compatible series or temporarily select mutually compatible branch revisions. Verify both sets with `RAPIDS_LIBS='rmm ucxx raft cuvs nvforest cuml' .devcontainer/launch.sh -d -c 13.3 -H rapids-conda -- ./ci/rapids/rapids-entrypoint.sh` and the equivalent command using `RAPIDS_LIBS='rmm ucxx raft cuvs cugraph wholegraph'`.

Copy this prompt into a coding agent

Repository: https://github.com/NVIDIA/cccl
Workflow run: https://github.com/NVIDIA/cccl/actions/runs/31476880502
Failure group: RAPIDS environment merges incompatible rapids-logger pins
Affected jobs:
- Build RAPIDS (optional) / rmm ucxx raft cuvs cugraph wholegraph: https://github.com/NVIDIA/cccl/actions/runs/31476880502/job/93732688509
- Build RAPIDS (optional) / rmm ucxx raft cuvs nvforest cuml: https://github.com/NVIDIA/cccl/actions/runs/31476880502/job/93732688511

Investigate the RAPIDS Conda environment failure for the CUDA 13.3 matrices `rmm ucxx raft cuvs nvforest cuml` and `rmm ucxx raft cuvs cugraph wholegraph`. Reproduce only environment generation/solving, inspect the cloned repositories' dependency YAML files to identify the sources of `rapids-logger==0.2.*,>=0.0.0a0` and `rapids-logger==0.3.*`, and confirm that a common logger series also leaves a CUDA 13.3-compatible cuML candidate. Implement the smallest fix by aligning dependency pins or selecting compatible RAPIDS branch revisions in the CI configuration; do not change the libcudacxx PR code to mask this dependency failure. Run focused environment-generation validation for both affected `RAPIDS_LIBS` sets and report the files changed and solver result.

Jobs:

• Build RAPIDS (optional) / rmm ucxx raft cuvs cugraph wholegraph
• Build RAPIDS (optional) / rmm ucxx raft cuvs nvforest cuml