coverage: expand resource, cache, and DLPack regression coverage by rluo8 · Pull Request #3060 · NVIDIA/cuda-python

rluo8 · 2026-10-09T08:33:14Z

Description

Expand cuda.core regression coverage with 35 new tests and additional cases in existing tests.

Extend graph-node reconstruction coverage for device memcpy, managed allocations, and peer-access metadata; validate invalid adjacency removal and kernel-node updates.
Cover managed-buffer validation, closed-buffer guards, VMM input validation, and stream workqueue resource metadata.
Exercise program-cache key generation, optional compiler discovery, disappearing cache entries, and file-permission handling.
Add DLPack coverage for CUDA-host/managed stream handling, dtype-width validation, missing ml_dtypes, and C exchange API exception propagation.
Cover NVVM compilation with a log sink, the NVRTC bundled-header version boundary, system-event validation, and kernel/array object behavior.

Changes are limited to tests and test comments.

Validation

All added tests passed on supported local test systems with Blackwell, Hopper, and Ampere GPUs.
Capability-dependent tests were skipped where unsupported.
All pre-commit checks passed.

Signed-off-by: Rui Luo <[email protected]>

coderabbitai · 2026-10-09T08:41:45Z

📝 Summary

Summary by CodeRabbit

Tests
- Expanded coverage for CUDA graph reconstruction, allocation metadata, node lifetimes, managed-memory operations, and context resources.
- Added checks for system-event validation, program compilation and caching, kernel behavior, tensor maps, opaque arrays, and DLPack interoperability.
- Verified error handling and object behavior across additional edge cases, including invalid arguments, closed buffers, cache failures, and unsupported operations.

This PR expands cuda_core tests across graph behavior, memory and resource APIs, program compilation and caching, kernel and array APIs, and DLPack validation. It adds checks for invalid inputs, closed objects, version gates, and filesystem race conditions.

Changes: Graph API Tests

Layer / File(s)	Summary
Graph node reconstruction `cuda_core/tests/graph/test_graph_definition_lifetime.py`	Reconstruction cases cover host and device memcpy nodes and allocation nodes with peer access or managed memory. Tests apply capability and build-version gates.
Graph mutation and kernel-node lifecycle `cuda_core/tests/graph/test_graph_definition_mutation.py`, `cuda_core/tests/graph/test_graph_node_update.py`	Tests check invalid adjacency removal, operations after source-node destruction, and updates to cooperative definition nodes.

Memory API Tests

Layer / File(s)	Summary
Managed-buffer attributes `cuda_core/tests/memory/test_managed_ops.py`	Tests check invalid values for the accessed-by set and attribute queries after buffer closure. Fixture comments describe coverage across both buffer sources.
Allocation argument validation `cuda_core/tests/test_memory.py`	Tests check that `modify_allocation` rejects a non-`Buffer` input and a regular `Buffer` with `TypeError`. The rejected buffer remains open.

CUDA Context and Resource Tests

Layer / File(s)	Summary
SM discovery and workqueue resources `cuda_core/tests/test_green_context.py`	Tests compare preferred co-scheduled SM discovery with native driver results and check primary-context stream workqueue properties against the device resource.

System Event Tests

Layer / File(s)	Summary
Event argument validation `cuda_core/tests/system/test_system_events.py`	Tests check the exceptions raised for an invalid event type and a non-container argument.

Program Compilation and Cache Tests

Layer / File(s)	Summary
Compilation and version-boundary behavior `cuda_core/tests/test_program.py`	Tests check NVVM compilation with a log sink and the NVRTC 13.2/13.3 bundled-headers boundary.
Program cache-key inputs `cuda_core/tests/test_program_cache.py`	Tests cover NVVM source normalization and fingerprint failures, PTX linker-selection failures, and cold nvJitLink imports.
File cache race and error paths `cuda_core/tests/test_program_cache.py`	Tests cover cache operations under filesystem errors and concurrent file disappearance, including size accounting and eviction.

Kernel Object Tests

Layer / File(s)	Summary
Kernel identity and representation `cuda_core/tests/test_module.py`	Tests compare kernels by handle, check hashing and inequality, and verify the representation prefix.

Tensor Map and Opaque Array Tests

Layer / File(s)	Summary
Tensor-map and opaque-array validation `cuda_core/tests/test_tensor_map.py`, `cuda_core/tests/test_texture_surface.py`	The tensor-map host-view test supplies an explicit interleave value. Opaque-array tests check format rejection, representation fields, and copy argument types.

DLPack Tests

Layer / File(s)	Summary
Capsule, dtype, and shape validation `cuda_core/tests/test_utils_dlpack.py`	Tests cover CUDA-host and CUDA-managed producer capsules, missing `ml_dtypes` behavior, and dtype lane-width shape validation.
C export error propagation `cuda_core/tests/test_utils_dlpack.py`	Tests check `BufferError` propagation from managed-tensor and borrowed-DLTensor export entry points.

Suggested reviewers: seberg

Priority: ⬇️ Low

Change: Other

Merge Risk: 🔵 Low · up to d2e96

The remaining concerns affect test reliability and regression detection, not demonstrated production behavior. The PR is mergeable with follow-up on those assertions.

- [ ] Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

coderabbitai

Actionable comments posted: 2

ℹ️ Review info

⚙️ Run configuration

Configuration used: Repository: NVIDIA/cuda-python/.coderabbit.yaml
Review profile: CHILL
Plan: Enterprise
Run ID: 0978b463-1b3d-4ef9-bdec-31e3dbe79d35

📥 Commits

Reviewing files that changed from the base of the PR and between a2b1e31 and a94c136.

📒 Files selected for processing (13)

cuda_core/tests/graph/test_graph_definition_lifetime.py
cuda_core/tests/graph/test_graph_definition_mutation.py
cuda_core/tests/graph/test_graph_node_update.py
cuda_core/tests/memory/test_managed_ops.py
cuda_core/tests/system/test_system_events.py
cuda_core/tests/test_green_context.py
cuda_core/tests/test_memory.py
cuda_core/tests/test_module.py
cuda_core/tests/test_program.py
cuda_core/tests/test_program_cache.py
cuda_core/tests/test_tensor_map.py
cuda_core/tests/test_texture_surface.py
cuda_core/tests/test_utils_dlpack.py

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 11 remain after this review.

github-actions · 2026-10-09T08:51:37Z

Doc Preview CI
Preview removed because the pull request was closed or merged.

Andy-Jost

Thanks, this looks good. CI ran every new test without a failure, and the gated tests ran too: the co-scheduling test passed on all six RTX PRO 6000 jobs with a CUDA 13 build. Two suggestions:

test_vmm_modify_allocation_rejects_non_buffer: object() stops at Cython's check of the buf: Buffer argument, so the method's own isinstance(buf, VirtualMemoryBuffer) check still has no test. That check is what makes the vmm_range() call valid. Could you add a case that passes a Buffer from another resource, such as init_cuda.memory_resource, and matches buf must be a VirtualMemoryBuffer?
test_from_dlpack_bfloat16_requires_ml_dtypes: could you use run_python_snippet from cuda_python_test_helpers.subprocess_runner, as the other snippet tests do? It runs the child in an empty directory and shows both output streams on failure, so -I is not needed.

Signed-off-by: Rui Luo <[email protected]>

coderabbitai

Actionable comments posted: 1

ℹ️ Review info

⚙️ Run configuration

Configuration used: Repository: NVIDIA/cuda-python/.coderabbit.yaml
Review profile: CHILL
Plan: Enterprise
Run ID: a184d412-858e-40fd-b9b1-84c6c80eeb61

📥 Commits

Reviewing files that changed from the base of the PR and between 2fe1140 and 1bab5d9.

📒 Files selected for processing (2)

cuda_core/tests/test_memory.py
cuda_core/tests/test_utils_dlpack.py

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 10 remain after this review.

Signed-off-by: Rui Luo <[email protected]>

rluo8 · 2026-10-13T08:57:14Z

Thanks @Andy-Jost . I have updated the tests and addressed both suggestions.

coderabbitai

🧹 Nitpick comments (1)

cuda_core/tests/test_program.py (1)

558-573: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

suggestion: Assert sink delivery.

The valid NVVM fixture produces an empty log, so this test only proves that logs= is accepted. A regression that ignores the sink can still return valid PTX and pass. Use a deterministic non-empty NVVM log case or a log-retrieval seam, then assert the sink content.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Repository: NVIDIA/cuda-python/.coderabbit.yaml
Review profile: CHILL
Plan: Enterprise
Run ID: 36d79019-5119-4b8c-af68-07f134fafab0

📥 Commits

Reviewing files that changed from the base of the PR and between 1bab5d9 and d2e96a6.

📒 Files selected for processing (8)

cuda_core/tests/graph/test_graph_definition_lifetime.py
cuda_core/tests/memory/test_managed_ops.py
cuda_core/tests/test_green_context.py
cuda_core/tests/test_memory.py
cuda_core/tests/test_module.py
cuda_core/tests/test_program.py
cuda_core/tests/test_tensor_map.py
cuda_core/tests/test_utils_dlpack.py

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 10 remain after this review.

Removed preview folders for the following PRs: - PR #2940 - PR #2993 - PR #3060 - PR #3082 - PR #3085 - PR #3086 - PR #3087 - PR #3091 - PR #3096

coverage: expand resource, cache, and DLPack regression coverage

a94c136

Signed-off-by: Rui Luo <[email protected]>

github-actions Bot added the cuda.core Everything related to the cuda.core module label Oct 9, 2026

rluo8 self-assigned this Oct 9, 2026

coderabbitai Bot reviewed Oct 9, 2026

View reviewed changes

Comment thread cuda_core/tests/test_memory.py

Comment thread cuda_core/tests/test_texture_surface.py

rluo8 requested review from Andy-Jost and mdboom October 9, 2026 11:02

rparolin added this to the cuda.core 1.3.0 milestone Oct 9, 2026

rparolin added the P0 High priority - Must do! label Oct 9, 2026

Merge branch 'main' into fix/core-test-coverage

2fe1140

Andy-Jost approved these changes Oct 12, 2026

View reviewed changes

test: cover non-VMM buffers and reuse subprocess runner

1bab5d9

Signed-off-by: Rui Luo <[email protected]>

coderabbitai Bot reviewed Oct 13, 2026

View reviewed changes

Comment thread cuda_core/tests/test_memory.py

rluo8 added 2 commits October 13, 2026 16:53

test: keep the DLPack subprocess helper import in the parent

9010d26

Signed-off-by: Rui Luo <[email protected]>

Merge branch 'main' into fix/core-test-coverage

d2e96a6

coderabbitai Bot reviewed Oct 13, 2026

View reviewed changes

Andy-Jost merged commit a864219 into NVIDIA:main Oct 13, 2026
108 checks passed

github-actions Bot pushed a commit that referenced this pull request Oct 14, 2026

Clean up PR preview folders for 9 closed/merged PRs

1abbb43

Removed preview folders for the following PRs: - PR #2940 - PR #2993 - PR #3060 - PR #3082 - PR #3085 - PR #3086 - PR #3087 - PR #3091 - PR #3096

Navigation Menu

Repository navigation

coverage: expand resource, cache, and DLPack regression coverage - #3060

coverage: expand resource, cache, and DLPack regression coverage#3060
Andy-Jost merged 5 commits into
NVIDIA:mainfrom
rluo8:fix/core-test-coverage

rluo8 commented Oct 9, 2026

Uh oh!

coderabbitai Bot commented Oct 9, 2026 •

edited

Loading

Summary by CodeRabbit

Uh oh!

coderabbitai Bot left a comment

Uh oh!

Uh oh!

Uh oh!

github-actions Bot commented Oct 9, 2026 •

edited

Loading

Uh oh!

Andy-Jost left a comment

Uh oh!

coderabbitai Bot left a comment

Uh oh!

Uh oh!

rluo8 commented Oct 13, 2026

Uh oh!

coderabbitai Bot left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

rluo8 commented Oct 9, 2026

Description

Uh oh!

coderabbitai Bot commented Oct 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

github-actions Bot commented Oct 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Andy-Jost left a comment

Choose a reason for hiding this comment

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

rluo8 commented Oct 13, 2026

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

coderabbitai Bot commented Oct 9, 2026 •

edited

Loading

github-actions Bot commented Oct 9, 2026 •

edited

Loading