fix a corner case of torch.arange #80758

yuguo68 · 2022-07-01T05:37:06Z

Stack from ghstack (oldest at bottom):

-> fix a corner case of torch.arange #80758

trying to fix #80589, see the two corner cases in the issue.
added the two cases in unit tests and add device to the tests.

Edit: the landed version only fixes the first issue in #80589

[ghstack-poisoned]

facebook-github-bot · 2022-07-01T05:37:12Z

🔗 Helpful links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/80758
📄 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 22ef735 (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.

trying to fix #80589, see the two corner cases in the issue. added the two cases in unit tests and add device to the tests. [ghstack-poisoned]

ghstack-source-id: 5af3b96 Pull Request resolved: #80758

ezyang · 2022-07-04T17:12:18Z

It would have been nice for the device parametrization to be separately stacked (no need to do now, I'll review as is)

ezyang · 2022-07-04T17:12:44Z

test/test_tensor_creation_ops.py

        torch.arange(1, 0, -1, out=res2)
        self.assertEqual(res1, res2, atol=0, rtol=0)
        torch.arange(1, 2, 1, out=res2)
        self.assertEqual(res1, res2, atol=0, rtol=0)

        # FloatTensor
-        res1 = torch.arange(0.6, 0.89, 0.1, out=torch.FloatTensor())
+        out = torch.tensor([], dtype=torch.float, device=device)
+        res1 = torch.arange(0.6, 0.89, 0.1, out=out)


good riddance, thx

ezyang · 2022-07-04T17:14:17Z

aten/src/ATen/native/RangeFactories.cpp

  AT_DISPATCH_ALL_TYPES_AND2(kHalf, kBFloat16, result.scalar_type(), "arange_cpu", [&]() {
    using accscalar_t = at::acc_type<scalar_t, false>;
    auto xstart = start.to<accscalar_t>();
    auto xend = end.to<accscalar_t>();
    auto xstep = step.to<accscalar_t>();

+    TORCH_CHECK(xstep > 0 || xstep < 0, "step must be nonzero");


Lol are there some floating point shenanigans why xstep != 0 is invalid here

preexisting condition

ezyang · 2022-07-04T17:17:04Z

test/test_tensor_creation_ops.py

+        r = torch.arange(0, 6, 3, device=device)
+        self.assertEqual(r.min(), 0)
+        self.assertEqual(r.max(), 3)
+        self.assertEqual(r.numel(), 2)


this is the new test

ezyang · 2022-07-04T17:17:21Z

test/test_tensor_creation_ops.py

+        r = torch.arange(0, -5, -2, device=device)
+        self.assertEqual(r.min(), -4)
+        self.assertEqual(r.max(), 0)
+        self.assertEqual(r.numel(), 3)


and this too, the test here is so bad woof

ezyang · 2022-07-04T17:17:52Z

onnx failure looks real

yuguo68 · 2022-07-05T16:19:53Z

onnx failure looks real

thanks for reviewing. Yes, it turns out in onnx we always cast the args (start, end, step) to desired dtype before computing size.
On master,

import torch
import numpy as np
>>> np.arange(1, 5.5, 1.5, dtype=np.int32) array([1, 2, 3], dtype=int32)
>>> np.arange(1, 5.5, 1.5, dtype=np.int64) array([1, 2, 3])
>>> torch.arange(1, 5.5, 1.5, dtype=torch.int32) tensor([1, 2, 3], dtype=torch.int32)
>>> torch.arange(1, 5.5, 1.5, dtype=torch.int64) tensor([1, 2, 3, 4])

This PR will change the last one to

>>> torch.arange(1, 5.5, 1.5, dtype=torch.int64) tensor([1, 2, 3])

but for onnx,

>>> onnx.arange(1, 5.5, 1.5, dtype=torch.int32) tensor([1, 2, 3, 4], dtype=torch.int32)
>>> onnx.arange(1, 5.5, 1.5, dtype=torch.int64) tensor([1, 2, 3, 4])

the failed unittest compare torch and onnx with int64, so it works on master but fails with this PR. We may need to change the implementation in torch.onnx to make it consistent.

trying to fix #80589, see the two corner cases in the issue. added the two cases in unit tests and add device to the tests. [ghstack-poisoned]

ghstack-source-id: bf36287 Pull Request resolved: #80758

ezyang · 2022-07-06T19:43:21Z

@pytorchbot merge -g

pytorchmergebot · 2022-07-06T19:44:34Z

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

github-actions · 2022-07-06T20:00:23Z

Hey @yuguo68.
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.

Summary: trying to fix #80589, see the two corner cases in the issue. added the two cases in unit tests and add device to the tests. Pull Request resolved: #80758 Approved by: https://github.com/ezyang Test Plan: contbuild & OSS CI, see https://hud.pytorch.org/commit/pytorch/pytorch/5e9136c24cf52d9da2e55ecb2049233c765f913b Reviewed By: mehtanirav Differential Revision: D37687413 Pulled By: yuguo68 fbshipit-source-id: befd1e384ff5d13272314bccac1e830b3c8db8de

fix two corner cases of torch.arange

27f0d73

[ghstack-poisoned]

facebook-github-bot added the cla signed label Jul 1, 2022

yuguo68 requested review from ngimel and ezyang July 1, 2022 05:45

lint on "fix two corner cases of torch.arange"

f992ed7

trying to fix #80589, see the two corner cases in the issue. added the two cases in unit tests and add device to the tests. [ghstack-poisoned]

yuguo68 added a commit that referenced this pull request Jul 1, 2022

fix two corner cases of torch.arange

454bf27

ghstack-source-id: 5af3b96 Pull Request resolved: #80758

ezyang reviewed Jul 4, 2022

View reviewed changes

ezyang approved these changes Jul 4, 2022

View reviewed changes

ezyang reviewed Jul 4, 2022

View reviewed changes

yuguo68 closed this Jul 5, 2022

yuguo68 reopened this Jul 5, 2022

Update on "fix two corner cases of torch.arange"

22ef735

trying to fix #80589, see the two corner cases in the issue. added the two cases in unit tests and add device to the tests. [ghstack-poisoned]

yuguo68 added a commit that referenced this pull request Jul 6, 2022

fix two corner cases of torch.arange

5732e92

ghstack-source-id: bf36287 Pull Request resolved: #80758

pytorchmergebot added the Merged label Jul 6, 2022

pytorchmergebot closed this in 5e9136c Jul 6, 2022

yuguo68 changed the title ~~fix two corner cases of torch.arange~~ fix a corner case of torch.arange Jul 6, 2022

yuguo68 added release notes: python_frontend python frontend release notes category topic: bug fixes topic category release notes labels Jul 6, 2022

yuguo68 removed the release notes label Jul 6, 2022

facebook-github-bot deleted the gh/yuguo68/2/head branch July 10, 2022 14:17

yanbing-j mentioned this pull request Mar 9, 2023

arange bug #95934

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix a corner case of torch.arange #80758

fix a corner case of torch.arange #80758

Uh oh!

yuguo68 commented Jul 1, 2022 •

edited

Loading

Uh oh!

facebook-github-bot commented Jul 1, 2022 •

edited

Loading

Uh oh!

ezyang commented Jul 4, 2022

Uh oh!

ezyang Jul 4, 2022

Uh oh!

ezyang Jul 4, 2022

Uh oh!

ezyang Jul 4, 2022

Uh oh!

ezyang Jul 4, 2022

Uh oh!

ezyang Jul 4, 2022

Uh oh!

ezyang commented Jul 4, 2022

Uh oh!

yuguo68 commented Jul 5, 2022 •

edited

Loading

Uh oh!

ezyang commented Jul 6, 2022

Uh oh!

pytorchmergebot commented Jul 6, 2022

Uh oh!

github-actions bot commented Jul 6, 2022

Uh oh!

Uh oh!

fix a corner case of torch.arange #80758

fix a corner case of torch.arange #80758

Uh oh!

Conversation

yuguo68 commented Jul 1, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

facebook-github-bot commented Jul 1, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful links

✅ No Failures (0 Pending)

Uh oh!

ezyang commented Jul 4, 2022

Uh oh!

ezyang Jul 4, 2022

Choose a reason for hiding this comment

Uh oh!

ezyang Jul 4, 2022

Choose a reason for hiding this comment

Uh oh!

ezyang Jul 4, 2022

Choose a reason for hiding this comment

Uh oh!

ezyang Jul 4, 2022

Choose a reason for hiding this comment

Uh oh!

ezyang Jul 4, 2022

Choose a reason for hiding this comment

Uh oh!

ezyang commented Jul 4, 2022

Uh oh!

yuguo68 commented Jul 5, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ezyang commented Jul 6, 2022

Uh oh!

pytorchmergebot commented Jul 6, 2022

Uh oh!

github-actions bot commented Jul 6, 2022

Uh oh!

Uh oh!

yuguo68 commented Jul 1, 2022 •

edited

Loading

facebook-github-bot commented Jul 1, 2022 •

edited

Loading

yuguo68 commented Jul 5, 2022 •

edited

Loading