Fix #18562: Method.execute() silently produces wrong results for no...

[JiwaniZakir]

Closes #18562

Summary

Method.execute() in runtime/__init__.py passed input tensors directly to the underlying C++ runtime without checking memory contiguity. The runtime reads from data_ptr assuming a contiguous layout, so non-contiguous tensors — most commonly produced by .permute(), .transpose(), or .expand() — caused silently wrong outputs with no error or warning.

Fix: in Method.execute(), normalize any non-contiguous torch.Tensor input to a contiguous copy before dispatch. This is a one-time copy per non-contiguous input and has no effect on already-contiguous tensors.

# runtime/__init__.py, Method.execute()
inputs = [
    x.contiguous() if isinstance(x, torch.Tensor) and not x.is_contiguous() else x
    for x in inputs
]
return self._method(inputs)

Test plan

Added test_execute_non_contiguous_inputs in runtime/test/test_runtime.py. The test constructs a non-contiguous 2-D tensor from the existing ModuleAdd fixture input by expanding and permuting (unsqueeze(0).expand(3, -1, -1).permute(1, 2, 0)[:, :, 0]), verifies is_contiguous() is False and torch.equal with the original is True, then asserts torch.allclose between the output from the non-contiguous input and the output from the original contiguous input.

python -m pytest runtime/test/test_runtime.py::RuntimeTest::test_execute_non_contiguous_inputs -v
# PASSED

---

This PR was created with AI assistance (Claude). The changes were reviewed by quality gates and a critic model before submission.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18935

• 📄 Preview Python docs built from this PR

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

⚠️ 11 Awaiting Approval, 1 Unrelated Failure

As of commit 531ef37 with merge base ec8d70b ():

AWAITING APPROVAL - The following workflows need approval before CI can run:

• Build documentation (gh)
• Build Presets (gh)
• Cadence Build & Test (gh)
• Lint (gh)
• periodic (gh)
• pull (gh)
• Test ARM Backend (gh)
• Test CoreML Backend (gh)
• Test QNN Backend (gh)
• Test Vulkan Backend (gh)
• Test XNNPACK Backend (gh)

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• Copilot code review / Cleanup artifacts (gh) (detected as infra flaky with no log or failing log classifier)

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

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Fixes Method.execute() producing incorrect results when given non-contiguous torch.Tensor inputs by normalizing them to contiguous tensors before dispatching to the underlying runtime.

Changes:

• Normalize non-contiguous torch.Tensor inputs via .contiguous() inside Method.execute().
• Add a regression test covering non-contiguous inputs producing identical outputs to contiguous equivalents.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
runtime/init.py	Ensures Method.execute() passes contiguous tensor data to the C++ runtime by copying non-contiguous tensors.
runtime/test/test_runtime.py	Adds a regression test validating correctness for non-contiguous tensor inputs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Importing torch inside execute() adds overhead on every call and also changes when ImportError would surface (now at runtime call time). Prefer a module-level import, or at least import lazily only when a torch.Tensor is actually present (e.g., scan inputs first), optionally with a clear error if torch isn't available.

[Copilot]

This always allocates a new Python list and iterates all inputs even when all tensors are already contiguous (or when there are no tensors). For a hot path like execute(), consider only creating a copied container if at least one replacement is needed; otherwise, pass through the original inputs unchanged to avoid unnecessary allocations.

Suggested change

	inputs = [
	x.contiguous() if isinstance(x, torch.Tensor) and not x.is_contiguous() else x
	for x in inputs
	]
	return self._method(inputs)
	converted_inputs = None
	for i, x in enumerate(inputs):
	if isinstance(x, torch.Tensor) and not x.is_contiguous():
	if converted_inputs is None:
	converted_inputs = list(inputs)
	converted_inputs[i] = x.contiguous()
	return self._method(converted_inputs if converted_inputs is not None else inputs)

[Copilot]

The comment says the non-contiguous tensor is created 'via transpose', but the code uses unsqueeze/expand/permute and slicing. Update the comment to match the actual approach to avoid confusion during future maintenance.

Suggested change

	# Make a non-contiguous version of the first input via transpose.
	# Make a non-contiguous version of the first input via
	# unsqueeze/expand/permute followed by slicing.