matrix_exp: Make sure _compute_linear_combinations result preserves dim of the input.

[nikitaved]

Fixes #80948.

[facebook-github-bot]

🔗 Helpful links

• 🧪  See artifacts and rendered test results at hud.pytorch.org/pr/81330
• 📄  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 our office hours

✅ No Failures (0 Pending)

As of commit 036e1ab (more details on the Dr. CI page):

Expand to see more

---

💚 💚 Looks good so far! There are no failures yet. 💚 💚

---

This comment was automatically generated by Dr. CI (expand for details).

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

[lezcano]

There was the pesky squeeze that was missing. Thank you @nikitaved!

Could you add a short test to make sure that we're fixing the issue?

[nikitaved]

The cases that trigger this warning are already in test_linalg.py. In fact, any valid input would have triggered them :)

[lezcano]

Yup, could you add one of those wrappers that make the test fail if a warning is thrown around those operations?

[nikitaved]

I seem to fail to capture this warning with OpInfo, for example:

================================================================================= FAILURES ==================================================================================
_______________________________________________________________ TestCommonCPU.test_out_warning_matrix_exp_cpu _______________________________________________________________
Unexpected success
______________________________________________________________ TestCommonCUDA.test_out_warning_matrix_exp_cuda ______________________________________________________________
Unexpected success
============================================================================= warnings summary ==============================================================================
...
  /home/nik/git/Quansight/pytorch/torch/testing/_internal/common_methods_invocations.py:953: UserWarning: An output with one or more elements was resized since it had shape [1, 3, 3], which does not match the required output shape [1, 1, 3, 3]. This behavior is deprecated, and in a future PyTorch release outputs will not be resized unless they have zero elements. You can explicitly reuse an out tensor t by resizing it, inplace, to zero elements with t.resize_(0). (Triggered internally at /home/nik/git/Quansight/pytorch/aten/src/ATen/native/Resize.cpp:17.)
    return self.op(*args, **kwargs)

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
========================================================================== short test summary info ==========================================================================
FAILED test/test_ops.py::TestCommonCPU::test_out_warning_matrix_exp_cpu
FAILED test/test_ops.py::TestCommonCUDA::test_out_warning_matrix_exp_cuda

This is done with expectedFailure. Am I missing something, @mruberry ?

[lezcano]

test_out_warning just tests for some warnings, like:

            # Case Zero: out= with the correct dtype and device, but the wrong shape
            #   Expected behavior: if nonempty, resize with a warning.

What I was suggesting was to add a short test in test_linalg.py using something like https://stackoverflow.com/questions/45671803/how-to-use-pytest-to-assert-no-warning-is-raised. Doing so for the matrix example in the issue would be enough.

[nikitaved]

Not sure this is strictly necessary, but decided to play it extra safe.

[lezcano]

Good stuff. Thank you Nik!

[lezcano]

@pytorchbot merge -g

[pytorchmergebot]

@pytorchbot successfully started a merge job. Check the current status here

[pytorchmergebot]

Merge failed due to Matched rule superuser, but PR #81330 was not reviewed yet by any of: nimin98, fbbradheintz, kavoor, lc0, shz117, ...
Raised by https://github.com/pytorch/pytorch/actions/runs/2658887357

[lezcano]

No one can say I didn't try. cc @mruberry

[mruberry]

Stamped!

[lezcano]

@pytorchbot merge -g

[pytorchmergebot]

@pytorchbot successfully started a merge job. Check the current status here

[github-actions]

Hey @nikitaved.
You've committed this PR, but it does not have both a 'release notes: ...' and 'topics: ...' label. Please add one of each to the PR. The 'release notes: ...' label should represent the part of PyTorch that this PR changes (fx, autograd, distributed, etc) and the 'topics: ...' label should represent the kind of PR it is (not user facing, new feature, bug fix, perf improvement, etc). The list of valid labels can be found here for the 'release notes: ...' and here for the 'topics: ...'.
For changes that are 'topic: not user facing' there is no need for a release notes label.

[malfet]

Merge failed due to Matched rule superuser, but PR #81330 was not reviewed yet by any of: nimin98, fbbradheintz, kavoor, lc0, shz117, ... Raised by https://github.com/pytorch/pytorch/actions/runs/2658887357

I think the problem here is how we parse merge rules. I.e. following line

pytorch/.github/merge_rules.json

Line 102 in 31142f5

"aten/src/ATen/native/**/*LinearAlgebra*",

would match any LinearAlgebra file in native/ subfolder, but not LinearAlgebra in the folder itself