Bugfix for doing negative padding #161639

skpark-rh · 2025-08-27T18:38:47Z

This bug fix introduces a fix that is consistent with the exception handling. Outlined in issue #161014, there is an edge case where the negative padding does not make the tensor size negative but still triggers the exception that the size is negative. The fix is simply adding new_dim >=0 to include the zero dim and letting the operator return an empty tensor.

In the PR I have added the edge case where the test will now check the negative padding where the dimension gets reduced to zero. But the sample is only for the constant type of padding. I would like some feedback if it is necessary to put the same sample on the reduce type as well.

This is my first PR to contribute to PyTorch and any help/feedback will be welcome! Thank you!

@malfet @manuelcandales @janeyx99 @ezyang

cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @pragupta @ezyang @msaroufim @dcci @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov @coconutruben @Lucaskabela

… with negative padding.

pytorch-bot · 2025-08-27T18:38:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/161639

📄 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 74a3823 with merge base 264e7f6 ():
💚 Looks good so far! There are no failures yet. 💚

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

linux-foundation-easycla · 2025-08-27T18:38:54Z

The committers listed above are authorized under a signed CLA.

✅ login: skpark-rh / name: Samuel Park (1c0563c, 9057042, b295e71, b6ea361, 3d7af38, c9f13f1, 554671f, 0eb565a, 56e210e, 3a29429, 6e31133, 640a033, 9071537, 74a3823, a545fcf, 5455f89)

skpark-rh · 2025-08-27T18:56:32Z

@pytorchbot label "topic: bug fixes"

… with negative padding.

… into bugfix/constant_pad_nd

skpark-rh · 2025-08-28T17:05:42Z

Hi @manuelcandales
So the only test that is failing doesn't seem to be related to the changes implemented in this fix. It is a torch distributed test... It is a NotImplementedError: Operator aten.linalg_inv_ex.default does not have a sharding strategy registered.

Skylion007 · 2025-08-29T17:20:37Z

aten/src/ATen/native/PadNd.cpp

        auto pad_idx = pad.size() - ((i + 1) * 2);
        auto new_dim = input_sizes[l_diff + i] + pad[pad_idx] + pad[pad_idx + 1];
-        TORCH_CHECK(new_dim > 0, "The input size ", input_sizes[l_diff + i], ", plus negative padding ",
+        TORCH_CHECK(new_dim >= 0, "The input size ", input_sizes[l_diff + i], ", plus negative padding ",


NIt: this should probably be TORCH_CHECK_VIEW

I don't see a TORCH_CHECK_VIEW in the repo... Could you point me to it so that I can see it and make appropriate changes?

ezyang · 2025-08-30T04:37:41Z

@manuelcandales please shepherd this to land, thanks!

skpark-rh · 2025-09-18T14:12:19Z

Hi just checking in with this PR. Any help landing this will be much appreciated.

@manuelcandales @ezyang

isuruf · 2025-09-19T12:50:38Z

@skpark-rh can you fix the failing test? The failing test was marked as xfail and we can remove it because of your fix.

skpark-rh · 2025-09-19T12:58:46Z

@isuruf Okay I can take a look at the failing test. The new test case, ((5, 3), (-1, -2, 1, 1)), that I put in torch/testing/_internal/common_methods_invocations.py may be causing the distributed tensor test to fail. Although I don't know how that is possible. Any pointers will be much appreciated!

isuruf · 2025-09-19T13:01:10Z

@skpark-rh the new test you added has nothing to do with it. The test test_dtensor_op_db_linalg_multi_dot_cpu_float32 failed with RuntimeError: The input size 0, plus negative padding 0 and 0 resulted in a negative output size, which is invalid. Check dimension 1 of your input. and therefore it was marked as a expected failure. With your PR, the test passes and therefore we just need to remove the xfail marker from the test.

skpark-rh · 2025-09-19T14:51:28Z

@isuruf Okay that makes sense. I removed the xfail linalg.multi_dot.

isuruf · 2025-09-19T15:33:17Z

@pytorchbot merge

pytorchmergebot · 2025-09-19T15:40:00Z

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

@malfet

Fixes pytorch#161014 This bug fix introduces a fix that is consistent with the exception handling. Outlined in issue pytorch#161014, there is an edge case where the negative padding does not make the tensor size negative but still triggers the exception that the size is negative. The fix is simply adding `new_dim >=0` to include the zero dim and letting the operator return an empty tensor. In the PR I have added the edge case where the test will now check the negative padding where the dimension gets reduced to zero. But the sample is only for the `constant` type of padding. I would like some feedback if it is necessary to put the same sample on the `reduce` type as well. This is my first PR to contribute to PyTorch and any help/feedback will be welcome! Thank you! @malfet @manuelcandales @janeyx99 @ezyang Pull Request resolved: pytorch#161639 Approved by: https://github.com/manuelcandales

@malfet

Fixes pytorch#161014 This bug fix introduces a fix that is consistent with the exception handling. Outlined in issue pytorch#161014, there is an edge case where the negative padding does not make the tensor size negative but still triggers the exception that the size is negative. The fix is simply adding `new_dim >=0` to include the zero dim and letting the operator return an empty tensor. In the PR I have added the edge case where the test will now check the negative padding where the dimension gets reduced to zero. But the sample is only for the `constant` type of padding. I would like some feedback if it is necessary to put the same sample on the `reduce` type as well. This is my first PR to contribute to PyTorch and any help/feedback will be welcome! Thank you! @malfet @manuelcandales @janeyx99 @ezyang Pull Request resolved: pytorch#161639 Approved by: https://github.com/manuelcandales

@malfet

Fixes pytorch#161014 This bug fix introduces a fix that is consistent with the exception handling. Outlined in issue pytorch#161014, there is an edge case where the negative padding does not make the tensor size negative but still triggers the exception that the size is negative. The fix is simply adding `new_dim >=0` to include the zero dim and letting the operator return an empty tensor. In the PR I have added the edge case where the test will now check the negative padding where the dimension gets reduced to zero. But the sample is only for the `constant` type of padding. I would like some feedback if it is necessary to put the same sample on the `reduce` type as well. This is my first PR to contribute to PyTorch and any help/feedback will be welcome! Thank you! @malfet @manuelcandales @janeyx99 @ezyang Pull Request resolved: pytorch#161639 Approved by: https://github.com/manuelcandales

skpark-rh added 3 commits August 26, 2025 16:56

Fixed bug where the new_dim check did not include 0 triggering assert…

1c0563c

… with negative padding.

Added edge case that checks for negative size in negative padding.

9057042

Updated the python ref file of constant_pad_nd to fix pytests.

3d7af38

skpark-rh requested a review from mruberry as a code owner August 27, 2025 18:38

pytorchbot added the open source label Aug 27, 2025

pytorch-bot bot added the topic: bug fixes topic category label Aug 27, 2025

skpark-rh added 6 commits August 28, 2025 02:34

Updated lazy tensor constant_pad_nd implementation.

b6ea361

Fixed bug where the new_dim check did not include 0 triggering assert…

c9f13f1

… with negative padding.

Added edge case that checks for negative size in negative padding.

b295e71

Updated the python ref file of constant_pad_nd to fix pytests.

3a29429

Updated lazy tensor constant_pad_nd implementation.

554671f

Merge branch 'bugfix/constant_pad_nd' of github.com:skpark-rh/pytorch…

a545fcf

… into bugfix/constant_pad_nd

skpark-rh requested review from a team, albanD, angelayi, avikchaudhuri, eqy, fmassa, jeffdaily, kulinseth, malfet, syed-ahmed, titaiwangms, tugsbayasgalan, ydwu4, zhxchen17 and zou3519 as code owners August 28, 2025 02:37

manuelcandales approved these changes Aug 28, 2025

View reviewed changes

manuelcandales removed module: inductor module: dynamo labels Aug 28, 2025

albanD removed their request for review August 28, 2025 15:54

Merge branch 'pytorch:main' into bugfix/constant_pad_nd

640a033

skpark-rh requested a review from manuelcandales August 29, 2025 13:24

Merge branch 'pytorch:main' into bugfix/constant_pad_nd

5455f89

Skylion007 reviewed Aug 29, 2025

View reviewed changes

Merge branch 'pytorch:main' into bugfix/constant_pad_nd

56e210e

skpark-rh added 2 commits September 19, 2025 14:00

Merge branch 'main' into bugfix/constant_pad_nd

0eb565a

Removed xfail "linalg.multi_dot" as the change fixes this test.

74a3823

pytorch-bot bot added the ciflow/trunk Trigger trunk jobs on your pull request label Sep 19, 2025

pytorchmergebot added the merging label Sep 19, 2025

pytorchmergebot added the Merged label Sep 19, 2025

pytorchmergebot closed this in df9a482 Sep 19, 2025

pytorchmergebot removed the merging label Sep 19, 2025

skpark-rh deleted the bugfix/constant_pad_nd branch September 19, 2025 21:06

Bugfix for doing negative padding #161639

Bugfix for doing negative padding #161639

Uh oh!

Conversation

skpark-rh commented Aug 27, 2025 • edited by pytorch-bot bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Aug 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/161639

✅ No Failures

Uh oh!

linux-foundation-easycla bot commented Aug 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

skpark-rh commented Aug 27, 2025

Uh oh!

skpark-rh commented Aug 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Skylion007 Aug 29, 2025

Choose a reason for hiding this comment

Uh oh!

skpark-rh Aug 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ezyang commented Aug 30, 2025

Uh oh!

skpark-rh commented Sep 18, 2025

Uh oh!

isuruf commented Sep 19, 2025

Uh oh!

skpark-rh commented Sep 19, 2025

Uh oh!

isuruf commented Sep 19, 2025

Uh oh!

skpark-rh commented Sep 19, 2025

Uh oh!

isuruf commented Sep 19, 2025

Uh oh!

pytorchmergebot commented Sep 19, 2025

Merge started

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

skpark-rh commented Aug 27, 2025 •

edited by pytorch-bot bot

Loading

pytorch-bot bot commented Aug 27, 2025 •

edited

Loading

linux-foundation-easycla bot commented Aug 27, 2025 •

edited

Loading

skpark-rh commented Aug 28, 2025 •

edited

Loading

skpark-rh Aug 29, 2025 •

edited

Loading