[libcu++] Implement P3104R6 Bit permutations by davebayer · Pull Request #10063 · NVIDIA/cccl

davebayer · 2026-07-22T07:13:47Z

No description provided.

coderabbitai · 2026-07-22T07:24:11Z

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

@coderabbitai resume to resume automatic reviews.
@coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

▶️ Resume reviews
🔍 Trigger review

📝 Walkthrough

Walkthrough

Changes

Adds cuda::std bit reversal, repetition, compression, and expansion utilities for unsigned integers. The umbrella header exports them, with constexpr and runtime tests across supported unsigned widths.

Bit operation utilities

Layer / File(s)	Summary
Primitive bit operations `libcudacxx/include/cuda/std/__bit/bit_reverse.h`, `libcudacxx/include/cuda/std/__bit/bit_repeat.h`, `libcudacxx/test/.../bit_reverse.pass.cpp`, `libcudacxx/test/.../bit_repeat.pass.cpp`	Adds constrained `bit_reverse` and `bit_repeat` APIs with type, `noexcept`, constexpr, runtime, and optional 128-bit coverage.
Masked bit transforms `libcudacxx/include/cuda/std/__bit/bit_compress.h`, `libcudacxx/include/cuda/std/__bit/bit_expand.h`, `libcudacxx/include/cuda/std/bit`, `libcudacxx/test/.../bit_compress.pass.cpp`, `libcudacxx/test/.../bit_expand.pass.cpp`	Adds mask-driven compression and expansion, exports the facilities through `cuda/std/bit`, and validates both implementations at runtime and compile time.

Suggested reviewers: ericniebler

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

coderabbitai

Actionable comments posted: 1

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 71b3dca6-5e9a-45ad-8e43-fd768a1f3e9a

📥 Commits

Reviewing files that changed from the base of the PR and between c2f0c18 and dedcdb7.

📒 Files selected for processing (9)

libcudacxx/include/cuda/std/__bit/bit_compress.h
libcudacxx/include/cuda/std/__bit/bit_expand.h
libcudacxx/include/cuda/std/__bit/bit_repeat.h
libcudacxx/include/cuda/std/__bit/bit_reverse.h
libcudacxx/include/cuda/std/bit
libcudacxx/test/libcudacxx/std/numerics/bit/bit.count/bit_compress.pass.cpp
libcudacxx/test/libcudacxx/std/numerics/bit/bit.count/bit_expand.pass.cpp
libcudacxx/test/libcudacxx/std/numerics/bit/bit.count/bit_repeat.pass.cpp
libcudacxx/test/libcudacxx/std/numerics/bit/bit.count/bit_reverse.pass.cpp

🚧 Files skipped from review as they are similar to previous changes (5)

libcudacxx/include/cuda/std/__bit/bit_reverse.h
libcudacxx/include/cuda/std/bit
libcudacxx/include/cuda/std/__bit/bit_repeat.h
libcudacxx/include/cuda/std/__bit/bit_expand.h
libcudacxx/include/cuda/std/__bit/bit_compress.h

fbusato

looks good. Left a couple of minor issues

fbusato · 2026-07-23T17:31:47Z

+#if _CCCL_HAS_INT128()
+  test<__uint128_t>();
+#endif // _CCCL_HAS_INT128()
+  return true;


note. Other bit operations also test for aliases. I think it makes more sense to only test for builtin types. Just a consistency note.

Yeah, I know, but I don't see the motivation behind testing the same types, just as aliases, this should cover all cases

miscco · 2026-07-24T06:54:55Z

+    const auto __n = ::cuda::std::countl_one(__mask_rev);
+
+    // Write __n consecutive bits.
+    const auto __segment = static_cast<_Tp>(__v & ::cuda::bitmask<_Tp>(0, __n));


Do we really need a bitmask here, or can we do the simple shift?

bitmask is more efficient on device, it produces a single BMSK instruction

miscco · 2026-07-24T07:10:34Z

+  const auto __pattern = static_cast<_Tp>(__v & ::cuda::bitmask<_Tp>(0, __n));
+  _Tp __ret            = __pattern;
+  for (int __i = __n; __i < __width; __i += __n)
+  {
+    __ret <<= __n;
+    __ret |= __pattern;
+  }
+  return __ret;
+}


Question: I am slightly worried about smaller than int sized integers. Wouldnt it be more efficient to promote to 32 bit and then cast back?

The masked bit operations are probably not efficient

Probably? I wanted to implement the simpler cases first and then we can optimize every algorithm separately

#10222, #10223, #10224

miscco · 2026-07-24T07:12:00Z

+    const auto __n = ::cuda::std::countl_one(__mask_rev);
+
+    // Write __n consecutive bits.
+    const auto __segment = static_cast<_Tp>(__v & ::cuda::bitmask<_Tp>(0, __n));


Question: Should we use the shl instruction on device?

At least clang knows some compiler builtins (__builtin_ia32_pext_di and __builtin_ia32_pext_si) for that https://www.mail-archive.com/cfe-commits@lists.llvm.org/msg706509.html

GCC knows _pext_u32 and _pext_u64 but only with -mbmi2 We might want to investigate if it is worth to optimize for clang-cuda here or at least add this as a followup

We will probably need the fallback for constexpr evaluation anyhow

miscco · 2026-07-24T07:14:14Z

+    return __v;
+  }
+
+  const auto __pattern = static_cast<_Tp>(__v & ::cuda::bitmask<_Tp>(0, __n));


Ditto we should be able to use compiler builtins for that

github-actions · 2026-07-30T08:30:42Z

🥳 CI Workflow Results

🟩 Finished in 2h 39m: Pass: 100%/120 | Total: 1d 22h | Max: 1h 48m | Hits: 100%/358474

See results here.

davebayer requested a review from a team as a code owner July 22, 2026 07:13

davebayer requested a review from ericniebler July 22, 2026 07:13

github-project-automation Bot added this to CCCL Jul 22, 2026

github-project-automation Bot moved this to Todo in CCCL Jul 22, 2026

cccl-authenticator-app Bot moved this from Todo to In Review in CCCL Jul 22, 2026

davebayer linked an issue Jul 22, 2026 that may be closed by this pull request

[FEA]: P3104R6 Bit permutations #10064

Closed

Jacobfaib reviewed Jul 22, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

davebayer force-pushed the bit_perms branch from c2f0c18 to dedcdb7 Compare July 23, 2026 15:15

coderabbitai Bot reviewed Jul 23, 2026

View reviewed changes

Comment thread libcudacxx/test/libcudacxx/std/numerics/bit/bit.count/bit_repeat.pass.cpp

Jacobfaib approved these changes Jul 23, 2026

View reviewed changes

davebayer force-pushed the bit_perms branch from dedcdb7 to e54008f Compare July 23, 2026 15:24

davebayer enabled auto-merge (squash) July 23, 2026 15:24

fbusato approved these changes Jul 23, 2026

View reviewed changes

davebayer disabled auto-merge July 23, 2026 17:35

This comment has been minimized.

Sign in to view

miscco reviewed Jul 24, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

[libcu++] Implement P3104R6 Bit permutations

7448f34

davebayer force-pushed the bit_perms branch from 3e81912 to 7448f34 Compare July 24, 2026 15:04

This comment has been minimized.

Sign in to view

disable noexcept test for old msvc and gcc

1af8735

This comment has been minimized.

Sign in to view

fix

1ac3452

davebayer merged commit 5c1d40f into NVIDIA:main Jul 30, 2026
138 of 140 checks passed

github-project-automation Bot moved this from In Review to Done in CCCL Jul 30, 2026

davebayer deleted the bit_perms branch July 30, 2026 08:33

Uh oh!

Conversation

davebayer commented Jul 22, 2026

Uh oh!

coderabbitai Bot commented Jul 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reviews paused

Walkthrough

Changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

fbusato left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

This comment has been minimized.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

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!

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.

This comment has been minimized.

github-actions Bot commented Jul 30, 2026

🥳 CI Workflow Results

🟩 Finished in 2h 39m: Pass: 100%/120 | Total: 1d 22h | Max: 1h 48m | Hits: 100%/358474

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

coderabbitai Bot commented Jul 22, 2026 •

edited

Loading