Add cuda::bit_ffs to <cuda/bit> by temujinkz · Pull Request #9609 · NVIDIA/cccl

temujinkz · 2026-06-26T11:06:10Z

cuda::bit_ffs(x) returns the 1-based position of the lowest set bit, or 0 if x is zero. The body is x == 0 ? 0 : countr_zero(x) + 1. countr_zero counts trailing zeros, so adding 1 turns it into a 1-based index. The x == 0 guard is the one case countr_zero doesn't define cleanly, and handling it is the whole reason this wrapper exists. It matches __builtin_ffs and CUDA's __ffs. It's constrained to unsigned integers, works on all widths including 128-bit, and is constexpr plus host/device because it just forwards to countr_zero.

Implements a type-safe find-first-set that returns 1 + the index of the least significant set bit, or 0 for a zero input, matching __builtin_ffs and CUDA's __ffs (unlike countr_zero). Adds tests and docs. Closes NVIDIA#6108 Signed-off-by: temujinkz <ttalkenov@gmail.com>

copy-pr-bot · 2026-06-26T11:06:14Z

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.

miscco · 2026-06-26T15:12:18Z

Thanks a lot for the PR, This already looks great, I have some nits

@miscco

Review feedback from @miscco and @fbusato on NVIDIA#9609: - route through __builtin_ffs / __builtin_ffsll on host and __ffs / __ffsll on device (constexpr fallback) so codegen matches __ffs / __ffsll - test: use assert instead of static_assert so the runtime path is exercised, mark the per-type helper void, [[maybe_unused]] the constant - test: add a concept check that bit_ffs rejects bool, signed, float, and enums - docs: add an all-bits-set example Signed-off-by: temujinkz <ttalkenov@gmail.com>

coderabbitai · 2026-06-27T08:41:38Z

📝 Walkthrough

Summary by CodeRabbit

New Features
- Added cuda::bit_ffs to <cuda/bit>, returning the 1-based position of the least significant set bit, or 0 for zero.
- Supports unsigned integer types, including 128-bit integers.
Documentation
- Added API reference documentation, availability details, usage guidance, and examples.
Tests
- Added coverage for supported types, compile-time and runtime evaluation, zero and multi-bit values, and invalid input types.

Walkthrough

Adds cuda::bit_ffs with host, device, and constexpr implementations, transitive inclusion from cuda/bit, extended API documentation, and tests for unsigned inputs and bit-position results.

Changes

Bit FFS API

Layer / File(s)	Summary
API surface `libcudacxx/include/cuda/__bit/bit_ffs.h`, `libcudacxx/include/cuda/bit`	Defines `cuda::bit_ffs` for unsigned integer types. It provides constexpr, host, and device implementations and exposes the API through `cuda/bit`.
Unsigned-input coverage `libcudacxx/test/libcudacxx/cuda/bit/bit_ffs.pass.cpp`	Checks rejected input categories and validates zero, single-bit, and multi-bit behavior across unsigned integer types at runtime and compile time.
Documentation and index `docs/libcudacxx/extended_api/bit.rst`, `docs/libcudacxx/extended_api/bit/bit_ffs.rst`	Documents the API, its 1-based least-significant-set-bit semantics, example usage, and CCCL/CUDA availability. The index links to the new page.

Suggested reviewers: miscco, gonidelis

Merge Risk: ⚪ Minimal · up to 8526d

The PR adds the new bit_ffs API and documentation with no actionable merge-blocking risk remaining; the formatting and documentation follow-ups are non-blocking.

_{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: 8617fc92-34c6-45a3-b206-feac28bfbaf5

📥 Commits

Reviewing files that changed from the base of the PR and between 8756dfc and bfb8bd1.

📒 Files selected for processing (5)

docs/libcudacxx/extended_api/bit.rst
docs/libcudacxx/extended_api/bit/bit_ffs.rst
libcudacxx/include/cuda/__bit/bit_ffs.h
libcudacxx/include/cuda/bit
libcudacxx/test/libcudacxx/cuda/bit/bit_ffs.pass.cpp

temujinkz · 2026-06-27T09:54:12Z

Thanks for the fast review, really appreciate it. Pushed an update that should cover everything:

@fbusato: bit_ffs now goes through __builtin_ffs / __builtin_ffsll on host and __ffs / __ffsll on device, with a constexpr fallback. I mirrored the host/device/constexpr dispatch from countr_zero so it stays consistent and matches the ffs/ffsll codegen.
@miscco: swapped the in-test static_asserts for assert so the runtime path actually runs, made the per-type helper return void, and marked the constant [[maybe_unused]].
@miscco: added a concept check so bit_ffs rejects bool, signed integers, float, and enums.
@miscco: added the all-bits-set example to the docs.

Let me know if the codegen looks right on your side, happy to tweak anything.

Pre-empt the same review feedback as bit_ffs (NVIDIA#9609): - implement bit_msb by forwarding to cuda::std::bit_width, which already lowers to the optimal clz based code - test: use assert instead of static_assert, per-type helper returns void, [[maybe_unused]] the constant, and add a concept check rejecting bool, signed, float, and enums - docs: add an all-bits-set example Signed-off-by: temujinkz <ttalkenov@gmail.com>

Per the IWYU coding guideline (and CodeRabbit), the test used numeric_limits and size_t through transitive includes. Add the direct headers. Signed-off-by: temujinkz <ttalkenov@gmail.com>

temujinkz · 2026-06-27T11:37:06Z

Good catch, fixed. Added <cuda/std/limits> and <cuda/std/cstddef> directly to the test instead of leaning on transitive includes.

temujinkz

went back through everything from the last round, should all be covered now. thanks for the thorough pass

- detect __builtin_ffs and __builtin_ffsll under separate conditions and drop the redundant clang/nvhpc checks; keep the gcc<10 fallback since gcc only gained __has_builtin in gcc 10, like the other bit headers - also undef the ffs builtins under nvrtc - rename __bit_ffs_impl_constexpr to __bit_ffs_impl_generic - take parameters by const value - drop the redundant intermediate casts and use cuda::std::uint32_t/uint64_t in the size dispatches - docs: bit_ffs ships in CUDA 13.5

Mark the signature __tile__ and include cuda_runtime_api.h in the example so it builds as written.

fbusato · 2026-07-15T17:02:57Z

/ok to test 5698ecb

Signed-off-by: temujinkz <ttalkenov@gmail.com>

griwes

This looks good, tests seem appropriate. Kinda wild we didn't have a native wrapper for __ffs yet.

miscco · 2026-08-14T06:23:41Z

/ok to test 8526d61

coderabbitai · 2026-08-14T06:23:42Z

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.

coderabbitai

🧹 Nitpick comments (2)

libcudacxx/include/cuda/__bit/bit_ffs.h (2)

35-41: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

suggestion: Update both #endif comments to repeat their complete opening conditions. Lines 37 and 41 omit || _CCCL_COMPILER(GCC, <, 10).

Based on learnings, each annotated #endif comment must repeat the exact verbatim condition from its corresponding #if.

Source: Learnings

119-124: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

suggestion: Convert the bit_ffs API comment to the required Doxygen form. Add //! @brief``, @param[in] __value, and `@return` documentation.

As per coding guidelines, documented functions must include //! @brief``, parameter documentation, and @return for non-void functions.

Source: Coding guidelines

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: cff4a5f2-d179-45ba-a298-83a5e503f268

📥 Commits

Reviewing files that changed from the base of the PR and between 3ae0e48 and 8526d61.

📒 Files selected for processing (5)

docs/libcudacxx/extended_api/bit.rst
docs/libcudacxx/extended_api/bit/bit_ffs.rst
libcudacxx/include/cuda/__bit/bit_ffs.h
libcudacxx/include/cuda/bit
libcudacxx/test/libcudacxx/cuda/bit/bit_ffs.pass.cpp

🚧 Files skipped from review as they are similar to previous changes (3)

libcudacxx/include/cuda/bit
docs/libcudacxx/extended_api/bit/bit_ffs.rst
docs/libcudacxx/extended_api/bit.rst

github-actions · 2026-08-14T07:03:20Z

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

Result: 0 regression row(s), 1 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	`0.629197`	`-0.629197`	`5.370743`	`4.741546`	Processing Header File: `libcudacxx/include/cuda/std/__cccl/prologue.h`	552

github-actions · 2026-08-14T08:02:01Z

🥳 CI Workflow Results

🟩 Finished in 1h 35m: Pass: 100%/117 | Total: 23h 27m | Max: 48m 03s | Hits: 98%/338174

See results here.

miscco · 2026-08-14T13:33:40Z

Thanks a lot for implementing this 🎉

Efficient bit fiddling is dear to my heart

temujinkz requested review from a team as code owners June 26, 2026 11:06

temujinkz requested a review from gonidelis June 26, 2026 11:06

github-project-automation Bot added this to CCCL Jun 26, 2026

temujinkz requested a review from griwes June 26, 2026 11:06

github-project-automation Bot moved this to Todo in CCCL Jun 26, 2026

cccl-authenticator-app Bot moved this from Todo to In Review in CCCL Jun 26, 2026

miscco requested changes Jun 26, 2026

View reviewed changes

github-project-automation Bot moved this from In Review to In Progress in CCCL Jun 26, 2026

miscco reviewed Jun 26, 2026

View reviewed changes

Comment thread docs/libcudacxx/extended_api/bit/bit_ffs.rst

fbusato reviewed Jun 27, 2026

View reviewed changes

Comment thread libcudacxx/include/cuda/__bit/bit_ffs.h Outdated

coderabbitai Bot reviewed Jun 27, 2026

View reviewed changes

Comment thread libcudacxx/test/libcudacxx/cuda/bit/bit_ffs.pass.cpp

bit_ffs: include <cuda/std/limits> and <cuda/std/cstddef> directly

9833415

Per the IWYU coding guideline (and CodeRabbit), the test used numeric_limits and size_t through transitive includes. Add the direct headers. Signed-off-by: temujinkz <ttalkenov@gmail.com>

temujinkz mentioned this pull request Jun 29, 2026

Add cuda::bit_msb to <cuda/bit> #9624

Merged

temujinkz commented Jul 9, 2026

View reviewed changes

fbusato requested changes Jul 10, 2026

View reviewed changes

temujinkz requested review from fbusato and miscco July 15, 2026 10:11

temujinkz added 2 commits July 15, 2026 15:27

Merge branch 'main' of https://github.com/NVIDIA/cccl into cuda-bit-ffs

b599837

bit_ffs: mirror docs feedback from the bit_msb review

5698ecb

Mark the signature __tile__ and include cuda_runtime_api.h in the example so it builds as written.

fbusato reviewed Jul 15, 2026

View reviewed changes

Comment thread libcudacxx/include/cuda/__bit/bit_ffs.h

Comment thread libcudacxx/include/cuda/__bit/bit_ffs.h Outdated

This comment has been minimized.

Sign in to view

bit_ffs: assume the result range and simplify the impl dispatch

0994742

Signed-off-by: temujinkz <ttalkenov@gmail.com>

temujinkz requested a review from fbusato July 17, 2026 08:09

fbusato approved these changes Jul 21, 2026

View reviewed changes

griwes approved these changes Aug 13, 2026

View reviewed changes

Merge branch 'main' into cuda-bit-ffs

8526d61

coderabbitai Bot reviewed Aug 14, 2026

View reviewed changes

miscco approved these changes Aug 14, 2026

View reviewed changes

github-project-automation Bot moved this from In Progress to In Review in CCCL Aug 14, 2026

miscco merged commit 2deb899 into NVIDIA:main Aug 14, 2026
145 checks passed

fbusato mentioned this pull request Sep 30, 2026

[libcu++] Implement cuda::ffs #6192

Closed

Conversation

temujinkz commented Jun 26, 2026

Uh oh!

copy-pr-bot Bot commented Jun 26, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

miscco commented Jun 26, 2026

Uh oh!

Uh oh!

Uh oh!

coderabbitai Bot commented Jun 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Walkthrough

Changes

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

temujinkz commented Jun 27, 2026

Uh oh!

temujinkz commented Jun 27, 2026

Uh oh!

temujinkz left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

fbusato commented Jul 15, 2026

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

griwes left a comment

Choose a reason for hiding this comment

Uh oh!

miscco commented Aug 14, 2026

Uh oh!

coderabbitai Bot commented Aug 14, 2026

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Aug 14, 2026

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

Direct file processing

Uh oh!

github-actions Bot commented Aug 14, 2026

🥳 CI Workflow Results

🟩 Finished in 1h 35m: Pass: 100%/117 | Total: 23h 27m | Max: 48m 03s | Hits: 98%/338174

Uh oh!

Uh oh!

miscco commented Aug 14, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

coderabbitai Bot commented Jun 27, 2026 •

edited

Loading