Fix libomp ABI bug by aliceb-nv · Pull Request #1593 · NVIDIA/cuopt

aliceb-nv · 2026-07-21T11:29:59Z

This PR adds a workaround resolving a buggy interaction between GCC's generated OMP helpers and LLVM's OpenMP runtime.
This was introduced as a latent issue in the Recursive RINS PR, and then exposed by the fast MPS PR which moved the wheel's runtime to LLVM OpenMP.

In more details:

The failure was caused by an LLVM libomp bug in its GCC GOMP_task compatibility layer: for an included OpenMP task (if(false)), libomp skipped GCC’s generated
firstprivate copy function, leaving a captured non-trivial C++ object unconstructed; the RINS sub-MIP path then read this invalid object and eventually threw
std::bad_array_new_length. The unrelated merge merely exposed the latent runtime-dependent bug in libomp-linked wheels; explicitly executing the included-task
path synchronously avoids the faulty compatibility path.

I will file a min repro and bug to LLVM later.

closes #1592

Description

Issue

Checklist

I am familiar with the Contributing Guidelines.
Testing
- New or existing tests cover these changes
- Added tests
- Created an issue to follow-up
- NA
Documentation
- The documentation is up to date with these changes
- Added new documentation
- NA

coderabbitai · 2026-07-21T11:34:56Z

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: 148614d1-61b4-4adc-baf8-2ca21a525777

📥 Commits

Reviewing files that changed from the base of the PR and between c21f7a8 and 67c5df3.

📒 Files selected for processing (1)

cpp/src/io/mps_parser.cpp

📝 Walkthrough

Walkthrough

Changes

Release alignment and package references

Layer / File(s)	Summary
Release workflow and package alignment `.github/workflows/*`, `RAPIDS_BRANCH`, `README.md`	Reusable workflows and the RAPIDS branch now use `release/26.08`; installation examples reference 26.08 package versions.

Fast MPS parsing and runtime behavior

Layer / File(s)	Summary
Quadratic MPS canonicalization `cpp/src/io/experimental_mps_fast/fast_parser.cpp`, `cpp/src/io/mps_parser.cpp`, `cpp/tests/linear_programming/experimental_mps_fast/fast_parser_edge_test.cpp`	Quadratic COO entries are validated and canonicalized, int64 specializations are instantiated, and the parser fixture includes an additional off-diagonal record.
RINS scheduling and regression coverage `cpp/src/branch_and_bound/branch_and_bound.cpp`, `skills/cuopt-developer/SKILL.md`, `python/cuopt/cuopt/tests/linear_programming/*`	RINS runs synchronously inside sub-MIPs and through an OpenMP task otherwise; related guidance and previously skipped tests are updated.
RMM type availability `cpp/src/pdlp/utilities/problem_checking.cuh`	The header directly includes the RMM `device_uvector` definition.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Suggested reviewers: rg20

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Out of Scope Changes check	⚠️ Warning	Several workflow, README, skill-doc, and parser/test-fixture edits are unrelated to the libomp workaround.	Split the unrelated release/version and parser changes into separate PRs so this change stays focused on the libomp fix and affected tests.
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title concisely describes the main change: fixing the libomp ABI bug.
Description check	✅ Passed	The description is directly about the libomp/OpenMP workaround and the swath1 crash it fixes.
Linked Issues check	✅ Passed	The PR addresses `#1592` by fixing the wheel crash and re-enabling the affected swath1 tests.

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests

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

github-actions · 2026-07-21T12:56:48Z

CI Test Summary

8 failed · 23 passed · 0 skipped

aliceb-nv · 2026-07-21T15:03:00Z

test failures unrelated. addressed in #1594

aliceb-nv · 2026-07-21T15:03:04Z

/merge

RAPIDS main moved to 26.10 after burndown; local builds were pulling rapids-cmake from RAPIDS main (26.10) while the conda env has 26.08 packages, breaking the build. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

rapids-cmake uses release/X.Y branch naming, not branch-X.Y. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

rapidsai/shared-workflows main moved to 26.10, causing wheel builds to use 26.10 CI containers. Pin to release/26.08 to match the RAPIDS_BRANCH. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

RMM defines device_uvector inside namespace RMM_NAMESPACE, which expands to `__attribute__((visibility("default"))) rmm` on GCC/Linux. A forward declaration in plain `namespace rmm {}` creates a second, distinct class visible to NVCC, causing '"rmm::device_uvector" is ambiguous' errors. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

…ith include" This reverts commit c9ae05a. Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

rmm 26.08.00a61 introduced inline namespace ABI versioning: device_uvector now lives in rmm::_RMM_26_8 (exposed as rmm::device_uvector via inline namespace). A plain-namespace forward declaration creates a second, distinct rmm::device_uvector, causing '"rmm::device_uvector" is ambiguous' errors in NVCC. Replace with the full header include. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

copy-pr-bot · 2026-07-22T07:17:42Z

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.

aliceb-nv · 2026-07-22T07:18:45Z

/ok to test c21f7a8

coderabbitai

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@cpp/src/io/mps_parser.cpp`:
- Around line 1720-1725: Instantiate check_symmetric_offdiagonal_pairs for
int64_t with both float and double value types alongside the existing
canonicalize_coo_matrix<int64_t, ...> explicit instantiations. Ensure these
emitted symbols satisfy the calls from parse_mps_fast_file<int64_t,
float/double> in the fast-parser translation unit.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: e5b06e46-be59-40c0-bb10-c11ddf171c90

📥 Commits

Reviewing files that changed from the base of the PR and between 84a6071 and c21f7a8.

📒 Files selected for processing (15)

.github/workflows/build.yaml
.github/workflows/multi_gpu_cpp_test.yaml
.github/workflows/pr.yaml
.github/workflows/test.yaml
.github/workflows/trigger-breaking-change-alert.yaml
RAPIDS_BRANCH
README.md
cpp/src/branch_and_bound/branch_and_bound.cpp
cpp/src/io/experimental_mps_fast/fast_parser.cpp
cpp/src/io/mps_parser.cpp
cpp/src/pdlp/utilities/problem_checking.cuh
cpp/tests/linear_programming/experimental_mps_fast/fast_parser_edge_test.cpp
python/cuopt/cuopt/tests/linear_programming/test_incumbent_callbacks.py
python/cuopt/cuopt/tests/linear_programming/test_lp_solver.py
skills/cuopt-developer/SKILL.md

💤 Files with no reviewable changes (1)

python/cuopt/cuopt/tests/linear_programming/test_lp_solver.py

🚧 Files skipped from review as they are similar to previous changes (2)

cpp/src/branch_and_bound/branch_and_bound.cpp
python/cuopt/cuopt/tests/linear_programming/test_incumbent_callbacks.py

aliceb-nv · 2026-07-22T08:00:50Z

/ok to test 67c5df3

aliceb-nv added this to the 26.08 milestone Jul 21, 2026

aliceb-nv requested review from a team as code owners July 21, 2026 11:30

aliceb-nv requested a review from ramakrishnap-nv July 21, 2026 11:30

aliceb-nv added bug Something isn't working non-breaking Introduces a non-breaking change labels Jul 21, 2026

aliceb-nv requested review from akifcorduk and rg20 July 21, 2026 11:30

akifcorduk approved these changes Jul 21, 2026

View reviewed changes

ramakrishnap-nv approved these changes Jul 21, 2026

View reviewed changes

chris-maes assigned aliceb-nv Jul 21, 2026

ramakrishnap-nv and others added 13 commits July 21, 2026 13:49

fix(build): use release/26.08 branch for rapids-cmake

c67db7a

rapids-cmake uses release/X.Y branch naming, not branch-X.Y. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

docs: update README install versions to 26.08

cb918ed

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

Revert "fix(build): replace rmm::device_uvector forward declaration w…

1862463

…ith include" This reverts commit c9ae05a. Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

ci: retrigger CI

d2000b3

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

ci: retrigger CI

8f61152

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

fix Qc test and added canonicalization

7050aad

style

97f14e3

added check symetrical off diagonal

7ba8b4a

fix libomp ABI bug

c21f7a8

aliceb-nv force-pushed the libomp-ifcopy-fix branch from 84a6071 to c21f7a8 Compare July 22, 2026 07:17

aliceb-nv requested a review from a team as a code owner July 22, 2026 07:17

aliceb-nv requested a review from bdice July 22, 2026 07:17

aliceb-nv removed the request for review from bdice July 22, 2026 07:18

coderabbitai Bot reviewed Jul 22, 2026

View reviewed changes

Comment thread cpp/src/io/mps_parser.cpp

missing instanciation

67c5df3

ramakrishnap-nv merged commit 87b32af into NVIDIA:main Jul 22, 2026
241 of 268 checks passed

Navigation Menu

Uh oh!

Fix libomp ABI bug - #1593

Fix libomp ABI bug#1593
ramakrishnap-nv merged 14 commits into
NVIDIA:mainfrom
aliceb-nv:libomp-ifcopy-fix

aliceb-nv commented Jul 21, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot commented Jul 21, 2026 •

edited

Loading

Walkthrough

Changes

❌ Failed checks (2 warnings)

Uh oh!

github-actions Bot commented Jul 21, 2026 •

edited

Loading

Uh oh!

aliceb-nv commented Jul 21, 2026

Uh oh!

aliceb-nv commented Jul 21, 2026

Uh oh!

copy-pr-bot Bot commented Jul 22, 2026

Uh oh!

aliceb-nv commented Jul 22, 2026

Uh oh!

coderabbitai Bot left a comment

Uh oh!

Uh oh!

aliceb-nv commented Jul 22, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Uh oh!

Conversation

aliceb-nv commented Jul 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Issue

Checklist

Uh oh!

coderabbitai Bot commented Jul 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

❌ Failed checks (2 warnings)

Uh oh!

github-actions Bot commented Jul 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CI Test Summary

Uh oh!

aliceb-nv commented Jul 21, 2026

Uh oh!

aliceb-nv commented Jul 21, 2026

Uh oh!

copy-pr-bot Bot commented Jul 22, 2026

Uh oh!

aliceb-nv commented Jul 22, 2026

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

aliceb-nv commented Jul 22, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

aliceb-nv commented Jul 21, 2026 •

edited

Loading

coderabbitai Bot commented Jul 21, 2026 •

edited

Loading

github-actions Bot commented Jul 21, 2026 •

edited

Loading