stft: Change require_complex warning to an error #49022

peterbell10 · 2020-12-08T15:59:17Z

BC-breaking note:

Previously torch.stft took an optional return_complex parameter that indicated whether the output would be a floating point tensor or a complex tensor. By default return_complex was False to be consistent with the previous behavior of torch.stft. This PR changes this behavior so return_complex is a required argument.

PR Summary:

Stack from ghstack:

Optimize torch.fft.fftn for real input with an r2c transform #46912 Optimize torch.fft.fftn for real input with an r2c transform
stft: Change require_complex warning to an error #49022 stft: Change require_complex warning to an error

Differential Revision: D25658906

[ghstack-poisoned]

dr-ci · 2020-12-08T16:54:48Z

💊 CI failures summary and remediations

As of commit a643614 (more details on the Dr. CI page):

2/3 failures possibly* introduced in this PR
- 1/2 non-CircleCI failure(s)
1/3 tentatively recognized as flaky ❄️
- Click here to rerun these jobs

🕵️ 1 new failure recognized by patterns

The following CI failures do not appear to be due to upstream breakages:

pytorch_libtorch_linux_xenial_cuda11_1_cudnn8_py3_gcc7_build (1/1)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

Dec 15 16:51:08 sccache: error: couldn't connect to server

Dec 15 16:51:08 +++ eval 'extract_trap_cmd ' 
Dec 15 16:51:08 ++++ extract_trap_cmd 
Dec 15 16:51:08 ++++ printf '%s\n' '' 
Dec 15 16:51:08 +++ printf '%s\n' cleanup 
Dec 15 16:51:08 ++ trap -- ' 
Dec 15 16:51:08 cleanup' EXIT 
Dec 15 16:51:08 ++ [[ pytorch-libtorch-linux-xenial-cuda11.1-cudnn8-py3-gcc7-build != *pytorch-win-* ]] 
Dec 15 16:51:08 ++ which sccache 
Dec 15 16:51:08 ++ sccache --stop-server 
Dec 15 16:51:08 Stopping sccache server... 
Dec 15 16:51:08 sccache: error: couldn't connect to server 
Dec 15 16:51:08 sccache: caused by: Connection refused (os error 111) 
Dec 15 16:51:08 ++ true 
Dec 15 16:51:08 ++ rm /var/lib/jenkins/sccache_error.log 
Dec 15 16:51:08 rm: cannot remove '/var/lib/jenkins/sccache_error.log': No such file or directory 
Dec 15 16:51:08 ++ true 
Dec 15 16:51:08 ++ [[ pytorch-libtorch-linux-xenial-cuda11.1-cudnn8-py3-gcc7-build == *rocm* ]] 
Dec 15 16:51:08 ++ SCCACHE_ERROR_LOG=/var/lib/jenkins/sccache_error.log 
Dec 15 16:51:08 ++ SCCACHE_IDLE_TIMEOUT=1200 
Dec 15 16:51:08 ++ RUST_LOG=sccache::server=error 
Dec 15 16:51:08 ++ sccache --start-server

❄️ 1 failure tentatively classified as flaky

but reruns have not yet been triggered to confirm:

pytorch_xla_linux_bionic_py3_6_clang9_test (1/1)

Step: "Run tests" (full log | diagnosis details | 🔁 rerun) ❄️

Dec 15 18:04:56 RuntimeError: tensorflow/compiler/xla/xla_client/xrt_local_service.cc:56 : Check failed: tensorflow::NewServer(server_def, &server_) == ::tensorflow::Status::OK() (Unknown: Could not start gRPC server vs. OK)

Dec 15 18:04:56   File "/opt/conda/lib/python3.6/site-packages/torch_xla-1.6-py3.6-linux-x86_64.egg/torch_xla/distributed/xla_multiprocessing.py", line 314, in _setup_replication 
Dec 15 18:04:56     device = xm.xla_device() 
Dec 15 18:04:56   File "/opt/conda/lib/python3.6/site-packages/torch_xla-1.6-py3.6-linux-x86_64.egg/torch_xla/core/xla_model.py", line 231, in xla_device 
Dec 15 18:04:56     devkind=devkind if devkind is not None else None) 
Dec 15 18:04:56   File "/opt/conda/lib/python3.6/site-packages/torch_xla-1.6-py3.6-linux-x86_64.egg/torch_xla/core/xla_model.py", line 136, in get_xla_supported_devices 
Dec 15 18:04:56     xla_devices = _DEVICES.value 
Dec 15 18:04:56   File "/opt/conda/lib/python3.6/site-packages/torch_xla-1.6-py3.6-linux-x86_64.egg/torch_xla/utils/utils.py", line 32, in value 
Dec 15 18:04:56     self._value = self._gen_fn() 
Dec 15 18:04:56   File "/opt/conda/lib/python3.6/site-packages/torch_xla-1.6-py3.6-linux-x86_64.egg/torch_xla/core/xla_model.py", line 18, in <lambda> 
Dec 15 18:04:56     _DEVICES = xu.LazyProperty(lambda: torch_xla._XLAC._xla_get_devices()) 
Dec 15 18:04:56 RuntimeError: tensorflow/compiler/xla/xla_client/xrt_local_service.cc:56 : Check failed: tensorflow::NewServer(server_def, &server_) == ::tensorflow::Status::OK() (Unknown: Could not start gRPC server vs. OK) 
Dec 15 18:04:57 Traceback (most recent call last): 
Dec 15 18:04:57   File "/var/lib/jenkins/workspace/xla/test/test_mp_all_gather.py", line 28, in <module> 
Dec 15 18:04:57     xmp.spawn(_mp_fn, args=()) 
Dec 15 18:04:57   File "/opt/conda/lib/python3.6/site-packages/torch_xla-1.6-py3.6-linux-x86_64.egg/torch_xla/distributed/xla_multiprocessing.py", line 394, in spawn 
Dec 15 18:04:57     start_method=start_method) 
Dec 15 18:04:57   File "/opt/conda/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 205, in start_processes 
Dec 15 18:04:57     while not context.join(): 
Dec 15 18:04:57   File "/opt/conda/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 160, in join 
Dec 15 18:04:57     exit_code=exitcode 
Dec 15 18:04:57 torch.multiprocessing.spawn.ProcessExitedException: process 3 terminated with exit code 17

ci.pytorch.org: 1 failed

Failed: pr/pytorch-linux-bionic-rocm3.9-py3.6

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.

See how this bot performed.

This comment has been revised 18 times.

[ghstack-poisoned]

mruberry · 2020-12-15T09:38:29Z

aten/src/ATen/native/SpectralOps.cpp

-  }
+  TORCH_CHECK(
+    return_complexOpt.has_value() || return_complex,
+    "stft requires the return_complex parameter be explicitly "


This is good. Do you think we should also take this opportunity to deprecate return_complex=False?

mruberry · 2020-12-15T09:39:48Z

torch/functional.py

-        Setting :attr:`return_complex` explicitly will be required in a future
-        PyTorch release. Set it to False to preserve the current behavior or
-        True to return a complex output.
+        From version 1.8.0, :attr:`return_complex` must be given explicitly for


This warning seems great. Same question about whether we should deprecate return_complex=False now, too?

Seems like we should? I don't think we intend to support float tensors as surrogates for complex tensors long term.

ghstack-source-id: 0c3cabe6a4bbd883273aa888e0bdb94a1687e1bc Pull Request resolved: pytorch#49022

[ghstack-poisoned]

peterbell10 · 2020-12-15T20:07:00Z

@mruberry I've rebased on viable/strict, deprecated return_complex=False and reworded the error message and doc warnings to point the user directly to return_complex=True.

Test failures look unrelated.

mruberry · 2020-12-15T21:09:56Z

aten/src/ATen/native/SpectralOps.cpp

+        "(which will be required in a future pytorch release). "
+      );
+
+    TORCH_WARN_ONCE(


This warning needs to be gated on the value actually being false, though. We don't want people who set return_complex=True to see it.

Oh wait... I understand how this will work now. OK. That's cool.

So the conditional only triggers if return_complex is unspecified or False and it needs to be specified. Then the check will capture the case where it's unspecified. If the check is hit this function will stop (since it threw an exception). If not, then the value is specified but it's False, so the warning is thrown.

Cool.

mruberry

Cool!

facebook-github-bot · 2020-12-16T21:23:20Z

@mruberry merged this pull request in 5874925.

mruberry · 2020-12-17T00:17:47Z

Hey @peterbell10, my mistake on landing this a little too swiftly. It looks like torchaudio has yet to update their stft calls. Let me ping them to do that before we land this so we don't break people using them.

mruberry · 2020-12-17T00:22:28Z

Torchaudio issue: pytorch/audio#1095

Summary: Pull Request resolved: pytorch#49022 Test Plan: Imported from OSS Reviewed By: ngimel Differential Revision: D25569586 Pulled By: mruberry fbshipit-source-id: 09608088f540c2c3fc70465f6a23f2aec5f24f85

stft: Change require_complex warning to an error

e7d1057

[ghstack-poisoned]

This was referenced Dec 8, 2020

Improve torch.fft n-dimensional transforms #46911

Closed

Use new FFT operators in stft #47601

Closed

Remove unused operator at::_fft_with_size #48905

Closed

facebook-github-bot added the cla signed label Dec 8, 2020

peterbell10 mentioned this pull request Dec 8, 2020

Optimize torch.fft.fftn for real input with an r2c transform #46912

Closed

pytorchbot added the open source label Dec 8, 2020

peterbell10 added 2 commits December 8, 2020 16:04

Update on "stft: Change require_complex warning to an error"

42ab446

[ghstack-poisoned]

Update on "stft: Change require_complex warning to an error"

bd4795e

[ghstack-poisoned]

peterbell10 added 2 commits December 8, 2020 22:49

Update on "stft: Change require_complex warning to an error"

c2682cd

[ghstack-poisoned]

Update on "stft: Change require_complex warning to an error"

971e0bc

[ghstack-poisoned]

mruberry reviewed Dec 15, 2020

View reviewed changes

peterbell10 added a commit to peterbell10/pytorch that referenced this pull request Dec 15, 2020

stft: Change require_complex warning to an error

743aee4

ghstack-source-id: 0c3cabe6a4bbd883273aa888e0bdb94a1687e1bc Pull Request resolved: pytorch#49022

Update on "stft: Change require_complex warning to an error"

a643614

[ghstack-poisoned]

mruberry reviewed Dec 15, 2020

View reviewed changes

mruberry approved these changes Dec 15, 2020

View reviewed changes

facebook-github-bot closed this in 5874925 Dec 16, 2020

facebook-github-bot added the Merged label Dec 16, 2020

mruberry added module: bc-breaking Related to a BC-breaking change module: deprecation labels Dec 17, 2020

mruberry reopened this Dec 17, 2020

facebook-github-bot closed this in 5c25f8f Dec 20, 2020

facebook-github-bot deleted the gh/peterbell10/34/head branch December 24, 2020 15:21

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

stft: Change require_complex warning to an error #49022

stft: Change require_complex warning to an error #49022

peterbell10 commented Dec 8, 2020 •

edited by mruberry

dr-ci bot commented Dec 8, 2020 •

edited

mruberry Dec 15, 2020

mruberry Dec 15, 2020

peterbell10 commented Dec 15, 2020

mruberry Dec 15, 2020

mruberry Dec 15, 2020

mruberry left a comment

facebook-github-bot commented Dec 16, 2020

mruberry commented Dec 17, 2020

mruberry commented Dec 17, 2020

stft: Change require_complex warning to an error #49022

stft: Change require_complex warning to an error #49022

Conversation

peterbell10 commented Dec 8, 2020 • edited by mruberry

dr-ci bot commented Dec 8, 2020 • edited

💊 CI failures summary and remediations

🕵️ 1 new failure recognized by patterns

pytorch_libtorch_linux_xenial_cuda11_1_cudnn8_py3_gcc7_build (1/1)

❄️ 1 failure tentatively classified as flaky

pytorch_xla_linux_bionic_py3_6_clang9_test (1/1)

ci.pytorch.org: 1 failed

mruberry Dec 15, 2020

Choose a reason for hiding this comment

mruberry Dec 15, 2020

Choose a reason for hiding this comment

peterbell10 commented Dec 15, 2020

mruberry Dec 15, 2020

Choose a reason for hiding this comment

mruberry Dec 15, 2020

Choose a reason for hiding this comment

mruberry left a comment

Choose a reason for hiding this comment

facebook-github-bot commented Dec 16, 2020

mruberry commented Dec 17, 2020

mruberry commented Dec 17, 2020

peterbell10 commented Dec 8, 2020 •

edited by mruberry

dr-ci bot commented Dec 8, 2020 •

edited