Add view_as_real, view_as_complex for complex tensors#39099
Closed
anjali411 wants to merge 38 commits intogh/anjali411/32/basefrom
Closed
Add view_as_real, view_as_complex for complex tensors#39099anjali411 wants to merge 38 commits intogh/anjali411/32/basefrom
anjali411 wants to merge 38 commits intogh/anjali411/32/basefrom
Conversation
[ghstack-poisoned]
This was referenced May 27, 2020
💊 CI failures summary and remediationsAs of commit 93e3cf5 (more details on the Dr. CI page): 💚 💚 Looks good so far! There are no failures yet. 💚 💚 This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group. This comment has been revised 238 times. |
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
[WIP] TODO: 1. add documentation 2. add tests 3. autograd code [ghstack-poisoned]
mruberry
reviewed
Jun 16, 2020
| x.real.sum().backward() | ||
| self.assertEqual(x.grad, torch.ones_like(x)) | ||
|
|
||
| # remove this test after gradcheck support is added for non-holomoprphic functions |
Collaborator
There was a problem hiding this comment.
what about a test where you manipulate both the real and imag parts and then backward through both those results?
Contributor
Author
There was a problem hiding this comment.
I don't see a case that this test covers which hasn't already been covered
mruberry
reviewed
Jun 16, 2020
mruberry
reviewed
Jun 16, 2020
mruberry
reviewed
Jun 16, 2020
|
|
||
| setattr(TestAutogradDeviceType, test_name, do_test) | ||
|
|
||
| class TestAutogradComplex(TestCase): |
Collaborator
There was a problem hiding this comment.
Should this class be made device generic so it runs on CPU and CUDA?
Contributor
Author
There was a problem hiding this comment.
Ithink autograd tests run for both cpu and cuda already
Collaborator
There was a problem hiding this comment.
These tests won't run on both but that's OK for now.
mruberry
reviewed
Jun 16, 2020
mruberry
reviewed
Jun 16, 2020
mruberry
reviewed
Jun 16, 2020
mruberry
reviewed
Jun 16, 2020
Differential Revision: [D22057886](https://our.internmc.facebook.com/intern/diff/D22057886) [ghstack-poisoned]
ezyang
reviewed
Jun 19, 2020
ezyang
reviewed
Jun 19, 2020
ezyang
reviewed
Jun 19, 2020
Differential Revision: [D22057886](https://our.internmc.facebook.com/intern/diff/D22057886) [ghstack-poisoned]
Differential Revision: [D22057886](https://our.internmc.facebook.com/intern/diff/D22057886) This PR adds the following: 1. `view_as_real`, `view_as_complex` for complex tensors 2. Adds the replay logic for these complex views that cause a change in the metadata (in this case- dtype): `as_strided` doesn't carry the tensor metadata information so this PR adds a replay logic similar to what `XLA` does for these view functions that cause a change in metadata. There are four possible scenarios concerning this PR: 1. view function that uses as_strided followed by a view function that uses replay (example. `.transpose(...)` followed by `view_as_real`) 2. view function that uses replay followed by a view function that uses as_strided (example. `view_as_real` followed by `.transpose(...)`) 3. view function that uses replay followed by another view function that uses replay (example. `view_as_real` followed by `view_as_complex`) 4. view function that uses as_strided followed by another view function that uses as_strided (this is what always happens for non-xla tensors rn) this covers case 1: https://github.com/pytorch/pytorch/pull/39099/files#diff-88d45e2b30fab214a6aa074fd30eb152R130 this covers case 2: https://github.com/pytorch/pytorch/pull/39099/files#diff-88d45e2b30fab214a6aa074fd30eb152R154 Case 3. is handled by existing XLA logic here: https://github.com/pytorch/pytorch/pull/39099/files#diff-88d45e2b30fab214a6aa074fd30eb152L120 [ghstack-poisoned]
mruberry
approved these changes
Jun 22, 2020
Collaborator
mruberry
left a comment
There was a problem hiding this comment.
Cool!
Just remember to update docs/source/torch.rst.
Differential Revision: [D22057886](https://our.internmc.facebook.com/intern/diff/D22057886) This PR adds the following: 1. `view_as_real`, `view_as_complex` for complex tensors 2. Adds the replay logic for these complex views that cause a change in the metadata (in this case- dtype): `as_strided` doesn't carry the tensor metadata information so this PR adds a replay logic similar to what `XLA` does for these view functions that cause a change in metadata. There are four possible scenarios concerning this PR: 1. view function that uses as_strided followed by a view function that uses replay (example. `.transpose(...)` followed by `view_as_real`) 2. view function that uses replay followed by a view function that uses as_strided (example. `view_as_real` followed by `.transpose(...)`) 3. view function that uses replay followed by another view function that uses replay (example. `view_as_real` followed by `view_as_complex`) 4. view function that uses as_strided followed by another view function that uses as_strided (this is what always happens for non-xla tensors rn) this covers case 1: https://github.com/pytorch/pytorch/pull/39099/files#diff-88d45e2b30fab214a6aa074fd30eb152R130 this covers case 2: https://github.com/pytorch/pytorch/pull/39099/files#diff-88d45e2b30fab214a6aa074fd30eb152R154 Case 3. is handled by existing XLA logic here: https://github.com/pytorch/pytorch/pull/39099/files#diff-88d45e2b30fab214a6aa074fd30eb152L120 [ghstack-poisoned]
albanD
approved these changes
Jun 22, 2020
Collaborator
albanD
left a comment
There was a problem hiding this comment.
Thanks Mike for the update.
Good to go from my side !
Contributor
|
@anjali411 merged this pull request in 8ec2ae9. |
laurentdupin
pushed a commit
to laurentdupin/pytorch
that referenced
this pull request
Apr 24, 2026
Summary: Pull Request resolved: pytorch#39099 Test Plan: Imported from OSS Differential Revision: D22057886 Pulled By: anjali411 fbshipit-source-id: bad5ba7097ba0dd13f2c549b2463094dee9afa14
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stack from ghstack:
Differential Revision: D22057886
This PR adds the following:
view_as_real,view_as_complexfor complex tensorsas_strideddoesn't carry the tensor metadata information so this PR adds a replay logic similar to whatXLAdoes for these view functions that cause a change in metadata. There are four possible scenarios concerning this PR:.transpose(...)followed byview_as_real)view_as_realfollowed by.transpose(...))view_as_realfollowed byview_as_complex)this covers case 1: https://github.com/pytorch/pytorch/pull/39099/files#diff-88d45e2b30fab214a6aa074fd30eb152R130
this covers case 2: https://github.com/pytorch/pytorch/pull/39099/files#diff-88d45e2b30fab214a6aa074fd30eb152R154
Case 3. is handled by existing XLA logic here: https://github.com/pytorch/pytorch/pull/39099/files#diff-88d45e2b30fab214a6aa074fd30eb152L120