`cuda::isclose()` by fbusato · Pull Request #9577 · NVIDIA/cccl

fbusato · 2026-06-24T01:29:02Z

Description

Comparing two floating-point values is a widely used functionality. Every library that use floating-point values need to compare them at some point. CCCL used it too for internal testing.

Python has math.isclose() https://peps.python.org/pep-0485/, and numpy.isclose.
Several C++ testing frameworks and libraries provide a similar utility, GoogleTest, Catch2, Boost.Math, Eigen.

This PR introduces 3 overloads for isclose() following implementation decisions of math.isclose()

bool isclose(T lhs, T rhs);
bool isclose(T lhs, T rhs, float relative_tolerance);
bool isclose(T lhs, T rhs, float relative_tolerance, T absolute_tolerance);

we could also think to add interfaces based on C++20 designed initializer.

Supported types: integers, (extended) floating-point, complex (std. cuda, cuda::std).

Todo: documentation (currently a draft)
Requires:

Support __float128 for isfinite() and isinf() #9576

coderabbitai · 2026-06-24T01:34:44Z

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

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: 548dfd63-db47-4b91-a13c-56bce3943b58

📥 Commits

Reviewing files that changed from the base of the PR and between 0fed909 and 5ed5965.

📒 Files selected for processing (1)

docs/libcudacxx/extended_api/numeric/isclose.rst

🚧 Files skipped from review as they are similar to previous changes (1)

docs/libcudacxx/extended_api/numeric/isclose.rst

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.

Summary

This PR adds cuda::isclose() (modeled after Python’s math.isclose() / cmath.isclose() as part of the work tracking #9576) for approximate equality of numeric values, providing scalar and complex overloads based on relative tolerance and (where supported) an absolute tolerance.

Changes

Implementation (`libcudacxx/include/cuda/__numeric/isclose.h`)

Added a new cuda/__numeric/isclose.h header implementing:
- Scalar overloads
  - isclose(lhs, rhs) (integers: exact equality; floating/extended: default relative tolerance)
  - isclose(lhs, rhs, rel_tol)
  - isclose(lhs, rhs, rel_tol, abs_tol) (integers/floating/extended)
- Complex overloads (__host__ __device__)
  - isclose(lhs, rhs, rel_tol, abs_tol) with participation constrained so abs_tol matches typename Complex::value_type
  - isclose(lhs, rhs, rel_tol)
  - isclose(lhs, rhs)
Key behavior:
- Default relative tolerance for extended/floating types is computed from numeric_limits<_Tp>::max_digits10 using a base-10 exponent derived from half that digit count.
- Validates:
  - rel_tol is in [0.0, 1.0]
  - abs_tol is finite and non-negative (with the signed-integer non-negativity check gated to signed integer types)
- Special-value handling:
  - Returns true on exact equality (and for complex, when real and imaginary parts are exactly equal).
  - Returns false if either operand/component is non-finite (NaN/Infinity cases).
- Approximate-equality criteria:
  - Floating-point / extended: abs(lhs - rhs) <= max(abs_tol, rel_tol * max(abs(lhs), abs(rhs)))
  - Integers: uses an overflow-safe unsigned absolute-difference helper, then applies the same tolerance threshold comparison.
  - Complex: otherwise compares Euclidean distance computed with hypot against max(abs_tol, rel_tol * max(|lhs|, |rhs|)), where |z| is computed via hypot.

Umbrella header integration (`libcudacxx/include/cuda/numeric`)

Added cuda/__numeric/isclose.h to libcudacxx/include/cuda/numeric.

Documentation

Added docs/libcudacxx/extended_api/numeric/isclose.rst documenting:
- all scalar and complex overloads
- tolerance parameter semantics, defaults (notably omitted abs_tol behavior), and constraints
- the NaN/Infinity rules and the approximate-equality criterion used for scalar/complex values
- a CUDA example kernel and main with assertions
Updated docs/libcudacxx/extended_api/numeric.rst:
- added the numeric/isclose page to the hidden toctree
- added a cuda::isclose entry row including availability metadata (CCCL 3.5.0, CUDA Toolkit 13.5).

Testing (`libcudacxx/test/libcudacxx/cuda/numeric/isclose/isclose.pass.cpp`)

Added a new pass test covering:
- compile-time checks for overload availability, return type, and noexcept
- runtime checks across floating-point, integral, and complex types (including NaN/Infinity)
- SFINAE/trait-based negative tests for invalid complex abs_tol combinations
- feature-gated runtime coverage for extended types (e.g., half/bfloat16) and host-only std::complex using existing target/feature macros.

Walkthrough

Adds cuda::isclose for scalar and complex approximate-equality checks, wires it into cuda/numeric, and adds matching documentation and coverage tests.

Changes

cuda::isclose numeric utility

Layer / File(s)	Summary
Documentation `docs/libcudacxx/extended_api/numeric.rst`, `docs/libcudacxx/extended_api/numeric/isclose.rst`	Adds the new reference page, documents overloads and semantics, and registers the page in the numeric navigation and availability table.
Internal helpers `libcudacxx/include/cuda/__numeric/isclose.h`	Defines the comparison-type selector, default relative tolerance, overflow-safe integer difference, and the floating-point, complex, and integer implementations.
Public overloads and header wiring `libcudacxx/include/cuda/__numeric/isclose.h`, `libcudacxx/include/cuda/numeric`	Adds the public scalar and complex overloads, closes the namespace and include guard, and includes the new header from `cuda/numeric`.
Test coverage `libcudacxx/test/libcudacxx/cuda/numeric/isclose/isclose.pass.cpp`	Adds compile-time and runtime tests for floating-point, integral, and complex overloads, including invalid `abs_tol` cases and target-gated runtime coverage.

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

Co-authored-by: Jacob Faibussowitsch <jacob.fai@gmail.com>

copy-pr-bot · 2026-06-26T19:48:51Z

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.

github-actions · 2026-06-27T00:24:55Z

😬 CI Workflow Results

🟥 Finished in 2h 20m: Pass: 99%/120 | Total: 2d 17h | Max: 2h 20m | Hits: 75%/587960

See results here.

fbusato added 3 commits June 22, 2026 12:48

draft

f5a8506

simplifications

f89bda2

unit test

3f07bfd

fbusato self-assigned this Jun 24, 2026

fbusato added the libcu++ For all items related to libcu++ label Jun 24, 2026

fbusato requested review from a team as code owners June 24, 2026 01:29

fbusato requested a review from gonidelis June 24, 2026 01:29

fbusato added this to CCCL Jun 24, 2026

fbusato requested a review from Jacobfaib June 24, 2026 01:29

github-project-automation Bot moved this to Todo in CCCL Jun 24, 2026

cccl-authenticator-app Bot moved this from Todo to In Review in CCCL Jun 24, 2026

fbusato commented Jun 24, 2026

View reviewed changes

Comment thread docs/libcudacxx/extended_api/numeric/isclose.rst

This comment has been minimized.

Sign in to view

davebayer changed the title ~~cuda::std::isclose()~~ cuda::isclose() Jun 24, 2026

Jacobfaib reviewed Jun 24, 2026

View reviewed changes

fbusato and others added 2 commits June 24, 2026 09:54

Update libcudacxx/include/cuda/__numeric/isclose.h

1b063b9

Co-authored-by: Jacob Faibussowitsch <jacob.fai@gmail.com>

a few improvements

d7f286b

This comment has been minimized.

Sign in to view

fbusato and others added 7 commits June 26, 2026 12:47

draft

31935cd

simplifications

c8f401a

unit test

765a138

Update libcudacxx/include/cuda/__numeric/isclose.h

1d50332

Co-authored-by: Jacob Faibussowitsch <jacob.fai@gmail.com>

a few improvements

c5657f4

improve test, docs

b22956c

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

5a86c05

fbusato requested review from a team as code owners June 26, 2026 19:48

fbusato requested review from miscco and rwgk June 26, 2026 19:48

delete temp

0ba5143

oleksandr-pavlyk reviewed Jun 26, 2026

View reviewed changes

Comment thread temp.build.yml Outdated

fbusato and others added 2 commits June 26, 2026 13:25

documentation fixes

1aa4485

Merge branch 'main' into isclose

0fed909

fbusato requested a review from davebayer June 26, 2026 20:27

doc: annotate exec space

5ed5965

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

`cuda::isclose()`#9577

`cuda::isclose()`#9577
fbusato wants to merge 16 commits into
NVIDIA:mainfrom
fbusato:isclose

fbusato commented Jun 24, 2026

Uh oh!

Uh oh!

coderabbitai Bot commented Jun 24, 2026 •

edited

Loading

Reviews paused

Uh oh!

This comment has been minimized.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

copy-pr-bot Bot commented Jun 26, 2026

Uh oh!

Uh oh!

github-actions Bot commented Jun 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

fbusato commented Jun 24, 2026

Description

Uh oh!

Uh oh!

coderabbitai Bot commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reviews paused

Summary

Changes

Implementation (libcudacxx/include/cuda/__numeric/isclose.h)

Umbrella header integration (libcudacxx/include/cuda/numeric)

Documentation

Testing (libcudacxx/test/libcudacxx/cuda/numeric/isclose/isclose.pass.cpp)

Walkthrough

Changes

Uh oh!

This comment has been minimized.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

copy-pr-bot Bot commented Jun 26, 2026

Uh oh!

Uh oh!

github-actions Bot commented Jun 27, 2026

😬 CI Workflow Results

🟥 Finished in 2h 20m: Pass: 99%/120 | Total: 2d 17h | Max: 2h 20m | Hits: 75%/587960

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

coderabbitai Bot commented Jun 24, 2026 •

edited

Loading

Implementation (`libcudacxx/include/cuda/__numeric/isclose.h`)

Umbrella header integration (`libcudacxx/include/cuda/numeric`)

Testing (`libcudacxx/test/libcudacxx/cuda/numeric/isclose/isclose.pass.cpp`)