[emitter] Ensure dim order and physical tensor memory match

[lucylq]

Summary

At the emitter, resolved tensors should be in contiguous (NCHW) or channels-last (NHWC) format, with the correct corresponding physical memory.

This PR checks that

1. If a tensor is not contiguous or channels-last, we call .contiguous on it.
2. After calling .contiguous, make sure the spec is updated with the new dim order.

Fixes a bug where channels-last tensors were made into NCHW, with the dim order not updated.

Test plan

python -m unittest executorch.exir.emit.test.test_emit.TestEmit.test_emit_channels_last_constant

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/16636

• 📄 Preview Python docs built from this PR

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

❌ 2 New Failures, 2 Unrelated Failures

As of commit 31335c6 with merge base 33974d5 ():

NEW FAILURES - The following jobs have failed:

• Lint / lintrunner / linux-job (gh)
  >>> Lint for exir/emit/test/test_emit.py:
• pull / android / run-emulator (gh)
  The process '/usr/bin/sh' failed with exit code 255

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

• pull / test-custom-ops-linux / linux-job (gh) (detected as infra flaky with no log or failing log classifier)

BROKEN TRUNK - The following job failed but was present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / test-samsung-models-linux / linux-job (gh) (trunk failure)
  ##[error]The operation was canceled.

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.

[Gasoonjia]

Why didn't we reset stride before?

[lucylq]

edit: set it here instead of after .contiguous() call to make sure it's correct.

[Gasoonjia]

ahh no no i mean before this PR there should be some incontiguous tensor being replaced by contiguous one by calling .contiguous(), and for those tensors they should also update spec. Wondering why we didn't do that before?

[lucylq]

Yeah not sure. We do update in tensor.py after calling contiguous though:

executorch/exir/tensor.py

Lines 189 to 209 in d58c8ee

	def from_tensor(cls, tensor: torch.Tensor, const: bool = False) -> TensorSpec:
	if const:
	# for non-contigous tensors, convert to a contiguous one
	tensor = tensor.contiguous()
	# Weights cannot be views during emission or serialization
	if tensor.nbytes != tensor.untyped_storage().nbytes():
	tensor = tensor.clone()
	spec = cls(
	dtype=tensor.dtype,
	shape=tensor.shape,
	layout=tensor.layout,
	const=const,
	is_sparse=tensor.is_sparse,
	)
	spec.stride = tensor.stride()
	spec.dim_order = dim_order_from_stride(spec.stride)
	spec.requires_grad = tensor.requires_grad
	spec.storage = tensor.untyped_storage() if const else None
	return spec

[Copilot]

Pull request overview

This PR fixes a bug where channels-last tensors were emitted with mismatched dim_order metadata and physical memory layout. The fix ensures that when processing constant tensors at emission time, both contiguous and channels-last formats are preserved correctly, and the TensorSpec metadata (stride and dim_order) is updated to match the actual tensor's memory layout.

Changes:

• Updated _emitter.py to check if tensors are either contiguous or channels_last, and update TensorSpec metadata accordingly
• Added dim_order_from_stride import to support computing correct dim_order from stride
• Added comprehensive test case validating channels-last constant tensor emission with verification of both dim_order metadata and physical storage layout

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
exir/emit/_emitter.py	Updated tensor contiguity check to allow channels-last format and ensure spec metadata is synchronized with actual tensor layout
exir/emit/test/test_emit.py	Added test validating channels-last constant tensors are emitted with correct dim_order and NHWC physical storage layout

---

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

[GregoryComer]

Thanks for fixing this! It might be worth adding one test for a constant tensor that has neither contiguous nor channels-last dim order. One motivating case was something like torch.randn(2, 16).transpose(1, 0).