Skip to content

Fix negative dimension handling in narrow_copy (#22969) - #22969

Merged
meta-codesync[bot] merged 1 commit into
pytorch:mainfrom
JakeStevens:export-D121015244
Sep 22, 2026
Merged

Fix negative dimension handling in narrow_copy (#22969)#22969
meta-codesync[bot] merged 1 commit into
pytorch:mainfrom
JakeStevens:export-D121015244

Conversation

@JakeStevens

@JakeStevens JakeStevens commented Sep 21, 2026 •
edited by meta-codesync Bot
Loading

Copy link
Copy Markdown
Contributor

Summary:

Resolves: #13556

Normalize negative dim values before validating narrow_copy arguments. This prevents valid negative dimensions from reaching tensor size checks unnormalized and causing a runtime assertion.

Adds regression coverage for copying with dim=-1.

Reviewed By: mergennachin

Differential Revision: D121015244

@pytorch-bot

pytorch-bot Bot commented Sep 21, 2026 •
edited
Loading

Copy link
Copy Markdown

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/22969

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

❌ 123 Cancelled Jobs, 2 Unrelated Failures

As of commit 46f544a with merge base bb2683b (image):

CANCELLED JOBS - The following jobs were cancelled. Please retry:

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

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

@meta-cla meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Sep 21, 2026
@meta-codesync

meta-codesync Bot commented Sep 21, 2026

Copy link
Copy Markdown
Contributor

@JakeStevens has exported this pull request. If you are a Meta employee, you can view the originating Diff in D121015244.

@JakeStevens JakeStevens linked an issue Sep 21, 2026 that may be closed by this pull request
ExecuTorch execution crashed for narrow_copy.default #13556
Closed
@github-actions

github-actions Bot commented Sep 21, 2026

Copy link
Copy Markdown

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.

@mergennachin

mergennachin commented Sep 21, 2026

Copy link
Copy Markdown
Contributor

The valid-negative-dimension fix works, but moving normalization before validation introduces a process-abort regression for out-of-range negative dimensions.

For input shape {1, 1, 1}, dim=-4, start=0, and length=1, the baseline returns Error::InvalidArgument. This change first converts dim to -1. check_narrow_copy_args() then accepts it because tensor_has_dim() accepts Python-style negative indices, and the subsequent in.size(-1) aborts:

tensor_impl.h:140: In function size(), assert failed (dim < dim_ && dim >= 0):
Dimension out of range (expected to be in range of [0, 2], but got -1

I reproduced this with the portable C++ kernel on macOS arm64, comparing head d3d29fdb50ef07e63d6e261b170700cd323ee032 against base bb2683b0049d57aaa27bca4117f2157ecce032fe, using the unchanged PR test source:

Test Base PR
New NegativeDimSupported SIGABRT Pass
Existing DimOutOfBoundDies Pass SIGABRT at dim=-4

The other nine tests pass on the PR. Despite its name, DimOutOfBoundDies expects a kernel error on the portable backend, so this abort is a test failure.

Please reject dimensions that remain negative after normalization, or validate the original dimension range before normalization. As a local diagnostic, changing the validation condition to the following made all 10 tests pass:

dim >= 0 && check_narrow_copy_args(in, dim, start, length, out)

@meta-codesync meta-codesync Bot changed the title Fix negative dimension handling in narrow_copy Fix negative dimension handling in narrow_copy (#22969) Sep 21, 2026
JakeStevens added a commit to JakeStevens/executorch that referenced this pull request Sep 21, 2026
@JakeStevens @facebook-github-bot
Fix negative dimension handling in narrow_copy (pytorch#22969)
4e7d7bf 
Summary:

Resolves: pytorch#13556

Normalize negative `dim` values before validating `narrow_copy` arguments. This prevents valid negative dimensions from reaching tensor size checks unnormalized and causing a runtime assertion.

Adds regression coverage for copying with `dim=-1`.

Differential Revision: D121015244
JakeStevens added a commit to JakeStevens/executorch that referenced this pull request Sep 21, 2026
@JakeStevens @facebook-github-bot
Fix negative dimension handling in narrow_copy (pytorch#22969)
dc8083c 
Summary:

Resolves: pytorch#13556

Normalize negative `dim` values before validating `narrow_copy` arguments. This prevents valid negative dimensions from reaching tensor size checks unnormalized and causing a runtime assertion.

Adds regression coverage for copying with `dim=-1`.

Reviewed By: mergennachin

Differential Revision: D121015244
JakeStevens added a commit to JakeStevens/executorch that referenced this pull request Sep 21, 2026
@JakeStevens @facebook-github-bot
Fix negative dimension handling in narrow_copy (pytorch#22969)
e26ff92 
Summary:

Resolves: pytorch#13556

Normalize negative `dim` values before validating `narrow_copy` arguments. This prevents valid negative dimensions from reaching tensor size checks unnormalized and causing a runtime assertion.

Adds regression coverage for copying with `dim=-1`.

Reviewed By: mergennachin

Differential Revision: D121015244
JakeStevens added a commit to JakeStevens/executorch that referenced this pull request Sep 21, 2026
@JakeStevens @meta-codesync
Fix negative dimension handling in narrow_copy (pytorch#22969)
46f544a 
Summary:
Pull Request resolved: pytorch#22969

Resolves: pytorch#13556

Normalize negative `dim` values before validating `narrow_copy` arguments. This prevents valid negative dimensions from reaching tensor size checks unnormalized and causing a runtime assertion.

Adds regression coverage for copying with `dim=-1`.

Reviewed By: mergennachin

Differential Revision: D121015244
@JakeStevens @facebook-github-bot
Fix negative dimension handling in narrow_copy (pytorch#22969)
67d384d 
Summary:

Resolves: pytorch#13556

Normalize negative `dim` values before validating `narrow_copy` arguments. This prevents valid negative dimensions from reaching tensor size checks unnormalized and causing a runtime assertion.

Adds regression coverage for copying with `dim=-1`.

Reviewed By: mergennachin

Differential Revision: D121015244
@meta-codesync
meta-codesync Bot merged commit 7dd3d9e into pytorch:main Sep 22, 2026
211 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@mergennachin mergennachin mergennachin approved these changes
@manuelcandales manuelcandales Awaiting requested review from manuelcandales manuelcandales is a code owner

Assignees

No one assigned

Labels

CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. meta-exported

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

ExecuTorch execution crashed for narrow_copy.default

2 participants

@JakeStevens @mergennachin