`cuda::device::warp_match_any` by fbusato · Pull Request #9243 · NVIDIA/cccl

fbusato · 2026-06-03T21:18:52Z

Description

The PR provides cuda::device::warp_match_any in a similar way of cuda::device::warp_match_all for completeness.
The main difference is that warp_match_any returns a mask instead of a bool value.

coderabbitai · 2026-06-03T21:26:15Z

Lost in the diff? Review this PR in Change Stack to follow the change map from intent to exact ranges.

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: 13b76f61-97d2-40ca-80d0-e5837894e2c8

📥 Commits

Reviewing files that changed from the base of the PR and between a17e799 and de51be0.

📒 Files selected for processing (1)

libcudacxx/test/libcudacxx/cuda/warp/warp_match_any.pass.cpp

Note: CodeRabbit is enabled on this repository as a convenience for maintainers
and contributors. Use your best judgment when considering its review comments and
suggestions — a suggested change may be inadequate, unnecessary, or safe to ignore.
Contributors are not expected to address every comment. Human reviews are what
ultimately matter for merging.

Overview

This PR introduces cuda::device::warp_match_any, a new warp-level utility function that returns a mask of lanes whose values match the calling lane. It mirrors the existing cuda::device::warp_match_all but provides a lane-value equality mask instead of a boolean result.

Changes

Core Implementation

New header: libcudacxx/include/cuda/__warp/warp_match_any.h
- Implements the templated warp_match_any function for SM_70+
- Supports any trivially copyable type _Tp
- Handles padding bit clearing when __builtin_clear_padding is available; otherwise requires bitwise-comparable types
- Processes data in 32-bit chunks using __match_any_sync intrinsic and intersects per-chunk results
- Adds an extern "C" device stub to signal unsupported pre-SM_70 targets

API Exposure

Updated libcudacxx/include/cuda/warp header to include and export the new functionality

Documentation

New documentation page: docs/libcudacxx/extended_api/warp/warp_match_any.rst
- Describes function signature, semantics, constraints, preconditions (SM ≥ 70, non-zero lane mask), undefined-behavior rules, performance notes, and references
- Includes example kernel and a note about padding-related compile errors unless __builtin_clear_padding is supported
Updated documentation: docs/libcudacxx/extended_api/warp/warp_match_all.rst
- Clarified memory ordering behavior and lane mask constraints
- Refined T type constraints regarding padding bit handling
- Updated example and references
Updated index: docs/libcudacxx/extended_api/warp.rst
- Added warp_match_any entry to the feature table (CCCL 3.5.0, CUDA 13.5)

Testing

New test file: libcudacxx/test/libcudacxx/cuda/warp/warp_match_any.pass.cpp
- Tests across multiple input types (integral widths, conditional __uint128_t, small structs/arrays)
- Validates behavior with different lane-mask patterns (prefix and strided masks)
- Includes tests for all-equal scenarios and alternating value patterns

suggestion:

Walkthrough

Adds cuda::device::warp_match_any: templated device function that serializes values into 32-bit chunks (optional padding clearing), calls __match_any_sync per chunk on SM_70+, intersects chunk masks, and returns the matching lane_mask. Includes tests, public header wiring, a new docs page, and related warp documentation updates.

Changes

warp_match_any Feature

Layer / File(s)	Summary
Core warp_match_any implementation `libcudacxx/include/cuda/__warp/warp_match_any.h`	Templated device function with input validation (trivially copyable, non-zero lane mask), chunk-based data serialization with optional `__builtin_clear_padding` support, per-chunk `__match_any_sync` invocation (SM_70+ with unsupported-before-SM_70 stub), and mask intersection returning final lane mask.
Warp module wiring `libcudacxx/include/cuda/warp`	Include directive for `cuda/__warp/warp_match_any.h` added to expose the new function.
Device validation tests `libcudacxx/test/libcudacxx/cuda/warp/warp_match_any.pass.cpp`	Device helpers (`make_low_mask`, `make_stride_mask`), templated test routines validating all-equal lanes and alternating even/odd values, kernel instantiation for integral types, conditional `__uint128_t`, and custom types (`char3`, `cuda::std::array<char,6>`), plus host entry point using `NV_DISPATCH_TARGET`.
warp_match_any documentation `docs/libcudacxx/extended_api/warp/warp_match_any.rst`	New user guide documenting signature, semantics, parameters, return value, constraints on `T`, execution preconditions (SM >= 70, non-zero lane mask), UB requirements, performance notes, references, and an example with Godbolt link.
Related documentation updates `docs/libcudacxx/extended_api/warp.rst`, `docs/libcudacxx/extended_api/warp/warp_match_all.rst`	Warp index `toctree` and feature table updated to include `warp_match_any`. `warp_match_all` docs clarified for memory-ordering, updated `T` constraint/padding wording, expanded `lane_mask` UB requirements, example comment change, and Godbolt link replacement.

Possibly Related PRs

NVIDIA/cccl#9192: Introduces the _CCCL_BUILTIN_CLEAR_PADDING macro and pattern for handling padding in warp-level operations, directly related to this PR's padding-handling logic.

Suggested Reviewers

ericniebler
gonidelis

Warning

There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure.

🔧 Infer (1.2.0)

libcudacxx/test/libcudacxx/cuda/warp/warp_match_any.pass.cpp

libcudacxx/test/libcudacxx/cuda/warp/warp_match_any.pass.cpp:14:10: fatal error: 'cuda/std/array' file not found
14 | #include <cuda/std/array>
| ^~~~~~~~~~~~~~~~
1 error generated.
Error: the following clang command did not run successfully:
/opt/infer-linux-x86_64-v1.2.0/lib/infer/facebook-clang-plugins/clang/install/bin/clang-18
@/tmp/coderabbit-infer/de51be0a903576c5b25d8c1cbfce8c8deeaf8147-bc434e2a457b67d9/tmp/clang_command_.tmp.33d91a.txt
++Contents of '/tmp/coderabbit-infer/de51be0a903576c5b25d8c1cbfce8c8deeaf8147-bc434e2a457b67d9/tmp/clang_command_.tmp.33d91a.txt':
"-cc1" "-load"
"/opt/infer-linux-x86_64-v1.2.0/lib/infer/infer/bin/../../facebook-clang-plugins/libtooling/build/FacebookClangPlugin.dylib"
"-add-plugin" "BiniouASTExporter" "-plugin-arg-BiniouASTExporter" "-"
"-plugin-arg-BiniouASTExporter" "PREPEND_CURRENT_DIR=1"
"-plugin-arg-BiniouASTExporter" "MAX_STRING_SIZE=65535" "-cc1" "-triple"
"x86_64-unknown-linux-gnu" "-

... [truncated 1152 characters] ...

nternal-isystem" "/usr/local/include" "-internal-isystem"
"/usr/lib/gcc/x86_64-linux-gnu/12/../../../../x86_64-linux-gnu/include"
"-internal-externc-isystem" "/usr/include/x86_64-linux-gnu"
"-internal-externc-isystem" "/include" "-internal-externc-isystem"
"/usr/include" "-Wno-ignored-optimization-argument" "-Wno-everything"
"-fdeprecated-macro" "-ferror-limit" "19" "-fgnuc-version=4.2.1"
"-fskip-odr-check-in-gmf" "-fcxx-exceptions" "-fexceptions"
"-D__GCC_HAVE_DWARF2_CFI_ASM=1" "-o"
"/tmp/coderabbit-infer/bc434e2a457b67d9/file.o" "-x" "c++"
"libcudacxx/test/libcudacxx/cuda/warp/warp_match_any.pass.cpp" "-O0"
"-fno-builtin" "-include"
"/opt/infer-linux-x86_64-v1.2.0/lib/infer/infer/bin/../lib/clang_wrappers/global_defines.h"
"-Wno-everything"

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

coderabbitai

🧹 Nitpick comments (1)

docs/libcudacxx/extended_api/warp/warp_match_any.rst (1)

36-37: 💤 Low value

suggestion: Constraints omit the is_bitwise_comparable requirement that the sibling warp_match_all doc lists for the no-__builtin_clear_padding case. Line 37 here describes padding but doesn't reference is_bitwise_comparable, while the implementation static_asserts it. Align with warp_match_all.rst line 37 for consistency.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 20e5a646-5b0a-4a6c-89fb-56027b96cc9e

📥 Commits

Reviewing files that changed from the base of the PR and between b336700 and d5c0bdb.

📒 Files selected for processing (6)

docs/libcudacxx/extended_api/warp.rst
docs/libcudacxx/extended_api/warp/warp_match_all.rst
docs/libcudacxx/extended_api/warp/warp_match_any.rst
libcudacxx/include/cuda/__warp/warp_match_any.h
libcudacxx/include/cuda/warp
libcudacxx/test/libcudacxx/cuda/warp/warp_match_any.pass.cpp

miscco · 2026-06-04T08:12:36Z

+  auto __data_copy = __data;
+  _CCCL_BUILTIN_CLEAR_PADDING(&__data_copy);
+  const auto __data_ptr = ::cuda::std::addressof(__data_copy);


Important: This introduces a needless copy. I believe this should only copy if is_bitwise_comparable_v is false

yes, I'm aware of this issue. However, is_bitwise_comparable_v also checks padding. I need to introduce another (internal) traits for that. I will open a second PR

Fine by me, although I would like to have an integer overload that does not do any of that and just forwards to the builtin

I don't think it is needed. Direct call and cuda::device::warp_match_any produce identical code as expected https://godbolt.org/z/sKaWz5dv1

jrhemstad · 2026-06-04T11:17:36Z

question: Should this be cuda::device::warp_match_any to be symmetrical with cuda::device::warp_match_all?

fbusato · 2026-06-04T16:22:46Z

question: Should this be cuda::device::warp_match_any to be symmetrical with cuda::device::warp_match_all?

sorry, only the title is wrong. The implementation is correct. Let me update it

Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>

… warp_match_any

github-actions · 2026-06-05T20:24:19Z

🥳 CI Workflow Results

🟩 Finished in 1h 12m: Pass: 100%/115 | Total: 1d 12h | Max: 1h 00m | Hits: 99%/336704

See results here.

fbusato added 4 commits June 3, 2026 13:24

implementation

53ab696

header

a5064cf

unit test

efeaa27

documentation

d5c0bdb

fbusato self-assigned this Jun 3, 2026

fbusato requested a review from a team as a code owner June 3, 2026 21:18

fbusato added the libcu++ For all items related to libcu++ label Jun 3, 2026

fbusato requested a review from a team as a code owner June 3, 2026 21:18

fbusato requested a review from gonidelis June 3, 2026 21:18

fbusato added this to CCCL Jun 3, 2026

fbusato requested a review from griwes June 3, 2026 21:18

github-project-automation Bot moved this to Todo in CCCL Jun 3, 2026

cccl-authenticator-app Bot moved this from Todo to In Review in CCCL Jun 3, 2026

coderabbitai Bot reviewed Jun 3, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

miscco reviewed Jun 4, 2026

View reviewed changes

fbusato changed the title ~~cuda::device::warp_any_match~~ cuda::device::warp_match_any Jun 4, 2026

fbusato and others added 4 commits June 4, 2026 09:24

Update libcudacxx/test/libcudacxx/cuda/warp/warp_match_any.pass.cpp

feb923e

Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>

Update libcudacxx/test/libcudacxx/cuda/warp/warp_match_any.pass.cpp

c0fe5cf

Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>

formatting

faa21ec

Merge branch 'warp_match_any' of https://github.com/fbusato/cccl into…

a17e799

… warp_match_any

fbusato requested a review from miscco June 4, 2026 16:35

fbusato enabled auto-merge (squash) June 4, 2026 16:35

This comment has been minimized.

Sign in to view

avoid launching a kernel

de51be0

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

`cuda::device::warp_match_any`#9243

`cuda::device::warp_match_any`#9243
fbusato wants to merge 9 commits into
NVIDIA:mainfrom
fbusato:warp_match_any

fbusato commented Jun 3, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot commented Jun 3, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot left a comment

Uh oh!

This comment has been minimized.

miscco Jun 4, 2026

Uh oh!

fbusato Jun 4, 2026 •

edited

Loading

Uh oh!

miscco Jun 5, 2026

Uh oh!

fbusato Jun 5, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jrhemstad commented Jun 4, 2026

Uh oh!

fbusato commented Jun 4, 2026

Uh oh!

This comment has been minimized.

github-actions Bot commented Jun 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

fbusato commented Jun 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Uh oh!

coderabbitai Bot commented Jun 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Overview

Changes

Core Implementation

API Exposure

Documentation

Testing

Walkthrough

Changes

Possibly Related PRs

Suggested Reviewers

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

miscco Jun 4, 2026

Choose a reason for hiding this comment

Uh oh!

fbusato Jun 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

miscco Jun 5, 2026

Choose a reason for hiding this comment

Uh oh!

fbusato Jun 5, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jrhemstad commented Jun 4, 2026

Uh oh!

fbusato commented Jun 4, 2026

Uh oh!

This comment has been minimized.

github-actions Bot commented Jun 5, 2026

🥳 CI Workflow Results

🟩 Finished in 1h 12m: Pass: 100%/115 | Total: 1d 12h | Max: 1h 00m | Hits: 99%/336704

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

fbusato commented Jun 3, 2026 •

edited

Loading

coderabbitai Bot commented Jun 3, 2026 •

edited

Loading

fbusato Jun 4, 2026 •

edited

Loading