[RiyaP2508]

Summary

Refactors test/test_multiprocessing.py to make device-agnostic tests run automatically on all available backends using instantiate_device_type_tests.

Previously, device coverage was achieved by manually duplicating tests (e.g. test_variable_sharing + test_cuda_variable_sharing), which meant out-of-tree backends had no coverage. This PR:

• Extracts a _MultiprocessingTestMixin with 4 shared helpers used by both test classes
• Creates TestMultiprocessingDeviceType with 8 device-generic test methods that replace 13 previously duplicated CPU/CUDA methods
• Fixes _test_autograd_sharing to derive device from var.device instead of hardcoding "cuda" if var.is_cuda else "cpu"
• Moves 3 @staticmethod child-process targets to module level for better pickle safety with mp.Process

All 13 CUDA-IPC-specific tests (events, IPC handles, _share_cuda_(), fork-poisoning, multi-GPU) remain unchanged in TestMultiprocessing with their original names and guards -- these use CUDA-only APIs with no accelerator-generic equivalent.

Tests moved to TestMultiprocessingDeviceType

New method	Replaces
test_simple_sharing	test_cuda_simple
test_empty_tensor_sharing	test_empty_tensor_sharing + test_empty_tensor_sharing_cuda
test_variable_sharing	test_variable_sharing + test_cuda_variable_sharing
test_parameter_sharing	test_parameter_sharing + test_cuda_parameter_sharing
test_integer_parameter_serialization	test_integer_parameter_serialization_cpu + _cuda
test_leaf_variable_sharing	test_leaf_variable_sharing (had internal device loop)
test_non_leaf_variable_sharing	test_non_leaf_variable_sharing (had internal device loop)
test_is_shared_accelerator	test_is_shared_cuda

Test plan

• CI should generate TestMultiprocessingDeviceTypeCPU and TestMultiprocessingDeviceTypeCUDA classes (visible with -v flag)
• All existing TestMultiprocessing tests pass unchanged
• On machines with XPU or OpenReg, additional device variants run automatically
• python test/test_multiprocessing.py -v to verify test discovery

cc: @fffrog @albanD @jbschlosser @mansiag05

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/184737

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit a9895a2 with merge base 6cd007b ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[RiyaP2508]

@pytorchbot rebase

[pytorchmergebot]

@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here

[pytorchmergebot]

Rebase failed due to Command git -C /home/runner/work/pytorch/pytorch rebase refs/remotes/origin/viable/strict pull/184737/head returned non-zero exit code 1

Rebasing (1/1)
Auto-merging test/test_multiprocessing.py
CONFLICT (content): Merge conflict in test/test_multiprocessing.py
error: could not apply 2d07daa9e58... Refactor test_multiprocessing.py: extract device-generic tests via instantiate_device_type_tests
hint: Resolve all conflicts manually, mark them as resolved with
hint: "git add/rm <conflicted_files>", then run "git rebase --continue".
hint: You can instead skip this commit: run "git rebase --skip".
hint: To abort and get back to the state before "git rebase", run "git rebase --abort".
hint: Disable this message with "git config set advice.mergeConflict false"
Could not apply 2d07daa9e58... # Refactor test_multiprocessing.py: extract device-generic tests via instantiate_device_type_tests Move 8 device-agnostic multiprocessing tests into TestMultiprocessingDeviceType, enabling automatic coverage on CPU, CUDA, XPU, and OpenReg/PrivateUse1 backends. Merge 13 duplicated CPU/CUDA test methods into device-parametrized equivalents. Fix _test_autograd_sharing to derive device from tensor instead of hardcoding CUDA. Move static child-process methods to module level for better pickle safety.

Raised by https://github.com/pytorch/pytorch/actions/runs/26795047228

[RiyaP2508]

@pytorchbot fix-lint

[pytorchmergebot]

Successfully applied lint patches in https://github.com/pytorch/pytorch/actions/runs/26795720228. Please pull locally before pushing more changes.

[jansel]

@claude review these changes

[claude]

Claude finished @jansel's task in 3m 55s —— View job

---

Review of PR #184737

• Read the PR diff and understand the changes
• Review code quality, correctness, and patterns
• Post review feedback

---

Summary

This PR refactors test/test_multiprocessing.py to introduce instantiate_device_type_tests for device-agnostic test coverage. The approach is sound: extract shared helpers into _MultiprocessingTestMixin, create TestMultiprocessingDeviceType with device-parameterized tests, and leave CUDA-IPC-specific tests in the original TestMultiprocessing class.

Issues

1. Duplicate tests after merge (Major)

The merge with main (8e97f6a) re-introduced the original test_variable_sharing, test_leaf_variable_sharing, test_non_leaf_variable_sharing, and test_cuda_variable_sharing methods into TestMultiprocessing (lines 1045-1098). These were supposed to be removed by this PR and replaced by the device-generic versions in TestMultiprocessingDeviceType (lines 507-550).

This means these tests now run twice — once in TestMultiprocessing with hardcoded device loops, and again via TestMultiprocessingDeviceTypeCPU / TestMultiprocessingDeviceTypeCUDA. This defeats the purpose of the refactoring and doubles test execution time for these cases.

The duplicated methods at lines 1045-1098 should be removed.

Fix this →

2. Unused imports (Minor)

The merge also brought in imports that are only used by the duplicate tests:

• skipIfRocm (line 27) — only used at line 1089 on the duplicate test_cuda_variable_sharing
• IS_LINUX (line 22) and TEST_WITH_SLOW (line 31) — used only by test_meta_simple skip decorator (line 1013), which came from main and is fine

If the duplicate tests are removed, skipIfRocm becomes unused and should be removed.

3. test_simple_sharing does not test test_cuda_simple's initialization (Minor)

The original test_cuda_simple (removed by this PR) had a line:

torch.cuda.FloatTensor([1])  # initialize CUDA outside of leak checker

The new test_simple_sharing in TestMultiprocessingDeviceType doesn't replicate this. For the CUDA path, this could potentially make leak_checker less reliable if CUDA initialization happens to open file descriptors inside the checker. This is likely a non-issue in practice (the leak checker's fd check is already disabled), but worth noting.

Design

The overall refactoring approach is clean and follows the established instantiate_device_type_tests pattern used elsewhere in PyTorch. Key positives:

• _MultiprocessingTestMixin cleanly separates shared logic
• _get_ctx(device) correctly handles spawn vs fork context selection
• _test_autograd_sharing now derives device from var.device instead of var.is_cuda — correct generalization
• Module-level functions for subprocess targets improve pickle safety
• allow_xpu=True on instantiate_device_type_tests is appropriate

Recommendation

The duplicate tests from the merge conflict resolution need to be removed before merging. Otherwise the refactoring goal of deduplication is not achieved.

---

[jansel]

address automated feedback from claude

[RiyaP2508]

address automated feedback from claude

Thanks for review @jansel , addressed them in latest push. PTAL:)

[RiyaP2508]

@pytorchbot merge

[pytorchmergebot]

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here