Fix unchecked CUDA calls in noexcept functions by lennartvoelz · Pull Request #1244 · NVIDIA/MatX

lennartvoelz · 2026-08-28T08:45:22Z

Adds MATX_CUDA_CHECK_NOEXCEPT:

checks the error state against cudaSuccess
on failure, logs the error
logging is guarded by try/catch, to allow safe usage in noexcept contexts (logging with std::format could throw)

Wraps bare CUDA API calls in either MATX_CUDA_CHECK or MATX_CUDA_CHECK_NOEXCEPT

… in noexcept contexts Adds MATX_CUDA_CHECK_NOEXCEPT: - checks the error state against cudaSuccess - on failure, logs the error - logging is guarded by try/catch, to allow safe usage in noexcept contexts (logging with std::format could throw)

copy-pr-bot · 2026-08-28T08:45:26Z

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.

…ry tracker

# Conflicts: # include/matx/core/allocator.h

Covers print.h, pybind.h, and tensor.h's prefetch helpers.

Covers cuda_executor_common.h's timing/profiling calls and distributed.h's device-guard and stream-cleanup destructors.

…forms/ Covers base_operator.h's copy/event-record paths and the solver transforms (cgsolve, chol, eig, fft, inverse, lu, qr, solve, svd).

greptile-apps · 2026-08-28T10:12:37Z

Greptile Summary

The PR adds checked CUDA-call handling, including non-throwing diagnostics for noexcept contexts, and converts the temporary stream and event lifetimes in iterative solver paths to RAII.

Adds MATX_CUDA_CHECK_NOEXCEPT for destructors and other noexcept functions.
Wraps previously unchecked CUDA runtime calls with the appropriate checking macro.
Introduces CUDA stream and event guards so successfully created handles are released during exception unwinding.
Updates cgsolve_impl and svdbpi_impl to use the new guards.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the previously reported exception-path resource leak is addressed by transferring each successfully created CUDA handle into RAII ownership before subsequent throwing operations.

Important Files Changed

Filename	Overview
include/matx/core/error.h	Adds a non-throwing CUDA error-check macro whose logging is protected against exceptions.
include/matx/core/resource_guard.h	Adds unique-pointer-based CUDA stream and event ownership with noexcept deleters.
include/matx/transforms/cgsolve.h	Transfers each successfully created convergence stream and event into RAII ownership before later checked operations can throw.
include/matx/transforms/svd/svd_cuda.h	Applies the same exception-safe ownership pattern to block-power-iteration SVD resources.

_{Reviews (3): Last reviewed commit: "Merge branch 'main' into fix/unchecked_c..." | Re-trigger Greptile}

The newly-added MATX_CUDA_CHECK calls in cgsolve_impl and svdbpi_impl can now throw between creating their temporary d2h stream/event and the end-of-function cleanup that used to destroy them, leaking both on any such throw. Replace the manual create/destroy with detail::CudaStreamGuard/CudaEventGuard (core/resource_guard.h), a shared unique_ptr-based RAII guard for opaque CUDA handles, so cleanup runs on any exception path.

cliffburdick · 2026-09-04T19:56:37Z

/build

lennartvoelz · 2026-09-10T18:41:27Z

Hi @cliffburdick, could you let me know which test is failing? I made sure no tests failed locally because of my changes, but some didn't run from the beginning because of the Jetson architecture, so it could be one of these cases I could not verify

cliffburdick · 2026-09-10T20:26:27Z

Hi @cliffburdick, could you let me know which test is failing? I made sure no tests failed locally because of my changes, but some didn't run from the beginning because of the Jetson architecture, so it could be one of these cases I could not verify

Hi @lennartvoelz, nothing is failing. It's just a DNS issue. I'll review and merge when ready.

cliffburdick · 2026-09-10T22:05:32Z

Hi @cliffburdick, could you let me know which test is failing? I made sure no tests failed locally because of my changes, but some didn't run from the beginning because of the Jetson architecture, so it could be one of these cases I could not verify

@lennartvoelz there is one unresolved comment from greptile.

lennartvoelz · 2026-09-12T15:36:57Z

@lennartvoelz there is one unresolved comment from greptile.

Hi @cliffburdick, I think the comment is stale. Should be fixed with the latest commit and the introduced resource guard 31c0b3c

cliffburdick · 2026-09-17T21:26:33Z

/build

coveralls · 2026-09-18T04:35:49Z

Coverage is 93.388% — lennartvoelz:fix/unchecked_cuda_calls into NVIDIA:main. No base build found for NVIDIA:main.

lennartvoelz added 7 commits August 28, 2026 10:56

Add MATX_CUDA_CHECK_NOEXCEPT macro around bare CUDA calls in the memo…

9ef52fd

…ry tracker

Add MATX_CUDA_CHECK around bare CUDA calls in cache.h

cfc86b7

Add MATX_CUDA_CHECK around bare cuda calls in make_sparse_tensor.h

61bf6d7

Merge branch 'main' into fix/unchecked_cuda_calls

f867806

# Conflicts: # include/matx/core/allocator.h

Add MATX_CUDA_CHECK around bare CUDA calls under core/

f631a62

Covers print.h, pybind.h, and tensor.h's prefetch helpers.

Add MATX_CUDA_CHECK around bare CUDA calls under executors/

f1bab6e

Covers cuda_executor_common.h's timing/profiling calls and distributed.h's device-guard and stream-cleanup destructors.

Add MATX_CUDA_CHECK around bare CUDA calls under operators/ and trans…

381af98

…forms/ Covers base_operator.h's copy/event-record paths and the solver transforms (cgsolve, chol, eig, fft, inverse, lu, qr, solve, svd).

lennartvoelz marked this pull request as ready for review August 28, 2026 10:09

greptile-apps Bot reviewed Aug 28, 2026

View reviewed changes

Comment thread include/matx/transforms/cgsolve.h

lennartvoelz and others added 2 commits August 28, 2026 13:40

Merge branch 'main' into fix/unchecked_cuda_calls

70b4a15

cliffburdick approved these changes Sep 18, 2026

View reviewed changes

cliffburdick merged commit bee23e5 into NVIDIA:main Sep 18, 2026
2 checks passed

Navigation Menu

Fix unchecked CUDA calls in noexcept functions - #1244

Fix unchecked CUDA calls in noexcept functions#1244
cliffburdick merged 10 commits into
NVIDIA:mainfrom
lennartvoelz:fix/unchecked_cuda_calls

lennartvoelz commented Aug 28, 2026

Uh oh!

copy-pr-bot Bot commented Aug 28, 2026

Uh oh!

greptile-apps Bot commented Aug 28, 2026 •

edited

Loading

Uh oh!

Uh oh!

cliffburdick commented Sep 4, 2026

Uh oh!

lennartvoelz commented Sep 10, 2026

Uh oh!

cliffburdick commented Sep 10, 2026

Uh oh!

cliffburdick commented Sep 10, 2026

Uh oh!

lennartvoelz commented Sep 12, 2026 •

edited

Loading

Uh oh!

cliffburdick commented Sep 17, 2026

Uh oh!

coveralls commented Sep 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

lennartvoelz commented Aug 28, 2026

Uh oh!

copy-pr-bot Bot commented Aug 28, 2026

Uh oh!

greptile-apps Bot commented Aug 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 5/5

Important Files Changed

Uh oh!

Uh oh!

cliffburdick commented Sep 4, 2026

Uh oh!

lennartvoelz commented Sep 10, 2026

Uh oh!

cliffburdick commented Sep 10, 2026

Uh oh!

cliffburdick commented Sep 10, 2026

Uh oh!

lennartvoelz commented Sep 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cliffburdick commented Sep 17, 2026

Uh oh!

coveralls commented Sep 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

greptile-apps Bot commented Aug 28, 2026 •

edited

Loading

lennartvoelz commented Sep 12, 2026 •

edited

Loading