Add host standard library detection #6244

davebayer · 2025-10-15T07:19:02Z

@ericniebler implemented some host standard library detection for using intrinsics for some simple functions such as move or forward.

I've improved the design a bit further, moved it to a separate header and added host standard library namespaces. The idea is that we could forward declare things like std::string instead of including the whole header to reduce compilation times.

The user would be required to include those headers if he wants to use APIs including these types.

Btw. _GLIBCXX_VERSION doesn't exist, I've changed it to __GLIBCXX__.

copy-pr-bot · 2025-10-15T07:19:05Z

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

miscco

Looks great, recapitulating the internal discussion for posterity

libcudacxx/include/cuda/std/__cccl/builtin.h

libcudacxx/include/cuda/std/__internal/namespaces.h

davebayer · 2025-10-15T08:35:00Z

/ok to test e1bcf4b

davebayer · 2025-10-15T08:46:34Z

/ok to test af71337

davebayer · 2025-10-15T08:52:24Z

/ok to test 0e45c13

libcudacxx/include/cuda/std/__cccl/builtin.h

libcudacxx/include/cuda/std/__internal/namespaces.h

davebayer · 2025-10-15T09:07:43Z

/ok to test 81588a9

davebayer · 2025-10-15T09:20:00Z

/ok to test 89b181a

davebayer · 2025-10-15T09:39:41Z

/ok to test e17232f

davebayer · 2025-10-15T11:08:06Z

libcudacxx/include/cuda/std/__complex/nvbf16.h

+template <class _CharT, class _Traits, class _Tp = float>
 ::std::basic_istream<_CharT, _Traits>&
 operator>>(::std::basic_istream<_CharT, _Traits>& __is, complex<__nv_bfloat16>& __x)
 {
-  ::std::complex<float> __temp;
+  ::std::complex<_Tp> __temp;


This is necessary to prevent errors about the type not being defined, but only forward declared

davebayer · 2025-10-15T11:08:23Z

libcudacxx/include/cuda/std/__complex/nvfp16.h

+  template <class _Up = value_type>
+  _CCCL_HOST operator ::std::complex<_Up>() const


davebayer · 2025-10-15T11:09:37Z

libcudacxx/include/cuda/std/string_view

 {
 #  if __cpp_lib_string_view >= 201606L
-  return __os << ::std::basic_string_view<_CharT>{__str};
+  return __os << ::std::basic_string_view<_CharT, ::std::char_traits<_CharT>>{__str};


Because we only forward declare the types, we don't have access to default template parameters, so we have to pass all of the template paramters here.

miscco · 2025-10-15T11:12:53Z

libcudacxx/include/cuda/std/__complex/nvbf16.h

+template <class _CharT, class _Traits, class _Tp = float>
 ::std::basic_ostream<_CharT, _Traits>&
 operator<<(::std::basic_ostream<_CharT, _Traits>& __os, const complex<__nv_bfloat16>& __x)
 {
-  return __os << complex<float>{__x};
+  return __os << complex<_Tp>{__x};
 }


This one should not be necessary because we are converting to a cuda::std::complex which is defined

miscco · 2025-10-15T11:13:52Z

libcudacxx/include/cuda/std/__complex/nvbf16.h

 operator>>(::std::basic_istream<_CharT, _Traits>& __is, complex<__nv_bfloat16>& __x)
 {
-  ::std::complex<float> __temp;
+  ::std::complex<_Tp> __temp;


I believe we could get away with the extended floating point types always casting to ::cuda::std::complex<float> and only do the smart thing for that

davebayer · 2025-10-16T06:44:08Z

libcudacxx/include/cuda/std/__cccl/builtin.h

-#    endif
-#  endif // defined(_GLIBCXX_VERSION) || defined(_LIBCPP_VERSION) || defined(_MSVC_STL_VERSION)
-#endif // defined(__cplusplus)
+// todo: re-enable std builtins


They were not working anyway, so I'd like to leave this for a different PR

I would love to know what is not woorking. Generally I would just replace the host STL detection and leave the rest in

davebayer · 2025-10-16T06:44:40Z

libcudacxx/include/cuda/std/__complex/nvfp16.h

  }
 };

-#  if !defined(_LIBCUDACXX_HAS_NO_LOCALIZATION) && !_CCCL_COMPILER(NVRTC)


Is never defined

miscco · 2025-10-16T12:30:01Z

libcudacxx/include/cuda/std/__cccl/builtin.h

-#    endif
-#  endif // defined(_GLIBCXX_VERSION) || defined(_LIBCPP_VERSION) || defined(_MSVC_STL_VERSION)
-#endif // defined(__cplusplus)
+// todo: re-enable std builtins


I would love to know what is not woorking. Generally I would just replace the host STL detection and leave the rest in

github-actions · 2025-10-16T13:37:47Z

🥳 CI Workflow Results

🟩 Finished in 6h 53m: Pass: 100%/127 | Total: 5d 18h | Max: 3h 20m | Hits: 42%/306565

See results here.

github-project-automation bot added this to CCCL Oct 15, 2025

github-project-automation bot moved this to Todo in CCCL Oct 15, 2025

cccl-authenticator-app bot moved this from Todo to In Progress in CCCL Oct 15, 2025

davebayer requested a review from miscco October 15, 2025 07:19

miscco reviewed Oct 15, 2025

View reviewed changes

libcudacxx/include/cuda/std/__cccl/builtin.h Outdated Show resolved Hide resolved

libcudacxx/include/cuda/std/__internal/namespaces.h Show resolved Hide resolved

miscco requested changes Oct 15, 2025

View reviewed changes

libcudacxx/include/cuda/std/__cccl/builtin.h Show resolved Hide resolved

libcudacxx/include/cuda/std/__internal/namespaces.h Outdated Show resolved Hide resolved

davebayer force-pushed the host_std_lib_detection branch from e17232f to 081bb70 Compare October 15, 2025 10:58

davebayer marked this pull request as ready for review October 15, 2025 10:58

davebayer requested review from a team as code owners October 15, 2025 10:58

davebayer requested review from elstehle and ericniebler October 15, 2025 10:58

cccl-authenticator-app bot moved this from In Progress to In Review in CCCL Oct 15, 2025

davebayer force-pushed the host_std_lib_detection branch from 081bb70 to dd0b986 Compare October 15, 2025 11:03

davebayer commented Oct 15, 2025

View reviewed changes

miscco requested changes Oct 15, 2025

View reviewed changes

github-project-automation bot moved this from In Review to In Progress in CCCL Oct 15, 2025

This comment has been minimized.

Sign in to view

davebayer force-pushed the host_std_lib_detection branch from 9231bee to ce1d9df Compare October 16, 2025 06:14

Add host standard library detection

02ff513

davebayer force-pushed the host_std_lib_detection branch from ce1d9df to 02ff513 Compare October 16, 2025 06:39

davebayer commented Oct 16, 2025

View reviewed changes

This comment has been minimized.

Sign in to view

davebayer requested a review from miscco October 16, 2025 12:27

miscco approved these changes Oct 16, 2025

View reviewed changes

github-project-automation bot moved this from In Progress to In Review in CCCL Oct 16, 2025

davebayer merged commit 13886f7 into NVIDIA:main Oct 16, 2025
271 of 275 checks passed

github-project-automation bot moved this from In Review to Done in CCCL Oct 16, 2025

		template <class _Up = value_type>
		_CCCL_HOST operator ::std::complex<_Up>() const

Add host standard library detection #6244

Add host standard library detection #6244

Conversation

davebayer commented Oct 15, 2025

Uh oh!

copy-pr-bot bot commented Oct 15, 2025

Uh oh!

miscco left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

davebayer commented Oct 15, 2025

Uh oh!

davebayer commented Oct 15, 2025

Uh oh!

davebayer commented Oct 15, 2025

Uh oh!

Uh oh!

Uh oh!

davebayer commented Oct 15, 2025

Uh oh!

davebayer commented Oct 15, 2025

Uh oh!

davebayer commented Oct 15, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

This comment has been minimized.

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Oct 16, 2025

🥳 CI Workflow Results

🟩 Finished in 6h 53m: Pass: 100%/127 | Total: 5d 18h | Max: 3h 20m | Hits: 42%/306565

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants