Replace async with non_blocking for Python 3.7 by goldsborough · Pull Request #4999 · pytorch/pytorch

goldsborough · 2018-02-01T22:12:33Z

Since async will become a keyword in Python 3.7, I replaced all occurences of the word async with non_blocking in our Python files. Proof:

$ find . -name '*.py' | xargs grep 'async\b'
./aten/src/ATen/copy_wrapper.py:if (async) {
./aten/src/ATen/copy_wrapper.py:if (async) {
./aten/src/ATen/copy_wrapper.py:Tensor & ${Type}::s_copy_(Tensor & self, const Tensor & src, bool async) const {
./torch/cuda/comm.py:    # TODO: if copying to CPU, allocate a pinned buffer, do async copies to it,

copy_wrapper.py generates C++ code, so the async will not be a problem. I also did not modify mentions of async in any other C++ files.

@ezyang @colesbury

Fixes #3431

ezyang · 2018-02-01T22:29:52Z

One question for the BC police: should we employ the following idiom to make legacy use of foo.cuda(async=True) keep working?

def _cuda(self, device=None, non_blocking=False, **kwargs):
    if 'async' in kwargs:
       assert not non_blocking
       non_blocking = kwargs['async']

EDIT: Actually, we should probably also check that async is the only member of kwargs, to guard against mispellings. dict.pop is a decent way to do this; then check if the dictionary is empty at the end.

ezyang · 2018-02-01T22:30:26Z

Thanks for undoing the C++ changes! The rest looks good.

goldsborough · 2018-02-01T22:31:15Z

Good point, I will add those compatibility fixes.

goldsborough · 2018-02-01T22:56:31Z

There's some trouble with CUDA tests, I am investigating.

goldsborough · 2018-02-02T00:09:26Z

@ezyang I had to rename async to non_blocking in torch/csrc/copy_utils.h because functions in there get wrapped and called from Python, which was causing errors. I then also renamed all other occurrences thereof in the C++ code so that:

Naming is consistent,
We don't run into similar trouble in the future.

I also don't know if some of the C++ code gets wrapped and exposed but just isn't called in the tests. Code also might get wrapped in the future, so I believe using non_blocking everywhere is better?

tools/autograd/templates/python_variable_methods.cpp

  HANDLE_TH_ERRORS
  static PythonArgParser parser({
-    "copy_(Tensor other, bool async=False)"
+    "copy_(Tensor other, bool non_blocking=False)"


torch/_utils.py

    """
+    if kwargs:
+        if len(kwargs) != 1 or 'async' not in kwargs:
+            raise ValueError("kwargs may only contain the key 'async'")


ezyang

YES!

goldsborough · 2018-02-02T05:34:08Z

I believe onnx-fb-universe tests are currently broken, so @ezyang if you are happy, let's merge it (assuming other tests will not fail)

Flake8 will produce different results on Python 2 and 3. Python 3.7 has __async__ as a reserved word pytorch#4999.

Summary: Flake8 will produce different results on Python 2 and 3. Python 3.7 has __async__ as a reserved word #4999. Pull Request resolved: #9953 Differential Revision: D9035415 Pulled By: soumith fbshipit-source-id: 8a46e028a2e20a7e3f6d90137020268d65a7cc64

XingangPan#2 and pytorch/pytorch#4999

Summary: Flake8 will produce different results on Python 2 and 3. Python 3.7 has __async__ as a reserved word pytorch#4999. Pull Request resolved: pytorch#9953 Differential Revision: D9035415 Pulled By: soumith fbshipit-source-id: 8a46e028a2e20a7e3f6d90137020268d65a7cc64

Fixes catalyst-team#3 The Dockerfiles are pip installing PyTorch=0.41 so this change is appropriate to match with pytorch/pytorch#4999

Modify the docs to match the changes made in pytorch#4999

Summary: goldsborough Modify the docs to match the changes made in #4999 Pull Request resolved: #12158 Differential Revision: D10103964 Pulled By: SsnL fbshipit-source-id: 1b8692da86aca1a52e8d2e6cea76a5ad1f71e058

__async__ is now a reserved word in Python 3.7 and later. To fix this pytorch/pytorch#4999 changed __cuda(async)__ to __cuda(non_blocking)__ so this PR tracks with that change. That fix landed in PyTourch 0.4.1.

Fixes CSAILVision#125 __async__ is a reserved word in Python 3.7 and later. To fix this pytorch/pytorch#4999 changed __cuda(async=True)__ to __cuda(non_blocking=True)__ so this PR tracks with that change which landed in PyTourch 0.4.1.

Fixes a __syntax error__ when running on Python >= 3.7 in alignment with pytorch/pytorch#4999 [flake8](http://flake8.pycqa.org) testing of https://github.com/ialhashim/DenseDepth on Python 3.7.1 $ __flake8 . --count --select=E9,F63,F72,F82 --show-source --statistics__ ``` ./PyTorch/train.py:58:78: E999 SyntaxError: invalid syntax depth = torch.autograd.Variable(sample_batched['depth'].cuda(async=True)) ^ 1 E999 SyntaxError: invalid syntax 1 ``` __E901,E999,F821,F822,F823__ are the "_showstopper_" [flake8](http://flake8.pycqa.org) issues that can halt the runtime with a SyntaxError, NameError, etc. These 5 are different from most other flake8 issues which are merely "style violations" -- useful for readability but they do not effect runtime safety. * F821: undefined name `name` * F822: undefined name `name` in `__all__` * F823: local variable name referenced before assignment * E901: SyntaxError or IndentationError * E999: SyntaxError -- failed to compile a file into an Abstract Syntax Tree

__async__ is now a reserved word in Python 3.7 and later. To fix this pytorch/pytorch#4999 changed __cuda(async=True)__ to __cuda(non_blocking=True)__ so this PR tracks with that change. That fix landed in PyTourch 0.4.1.

__async__ is a reserved word in Python 3.7 and later. To fix this pytorch/pytorch#4999 changed __cuda(async=True)__ to __cuda(non_blocking=True)__ so this PR tracks with that change. That fix landed in PyTourch 0.4.1.

Using __async__ in this way is a Syntax Error in Python >= 3.7. This change tracks with the changes made in pytorch/pytorch#4999

Fixes CSAILVision#125 __async__ is a reserved word in Python 3.7 and later. To fix this pytorch/pytorch#4999 changed __cuda(async=True)__ to __cuda(non_blocking=True)__ so this PR tracks with that change which landed in PyTourch 0.4.1.

Fixes a __syntax error__ when running on Python >= 3.7 in alignment with pytorch/pytorch#4999 [flake8](http://flake8.pycqa.org) testing of https://github.com/ialhashim/DenseDepth on Python 3.7.1 $ __flake8 . --count --select=E9,F63,F72,F82 --show-source --statistics__ ``` ./PyTorch/train.py:58:78: E999 SyntaxError: invalid syntax depth = torch.autograd.Variable(sample_batched['depth'].cuda(async=True)) ^ 1 E999 SyntaxError: invalid syntax 1 ``` __E901,E999,F821,F822,F823__ are the "_showstopper_" [flake8](http://flake8.pycqa.org) issues that can halt the runtime with a SyntaxError, NameError, etc. These 5 are different from most other flake8 issues which are merely "style violations" -- useful for readability but they do not effect runtime safety. * F821: undefined name `name` * F822: undefined name `name` in `__all__` * F823: local variable name referenced before assignment * E901: SyntaxError or IndentationError * E999: SyntaxError -- failed to compile a file into an Abstract Syntax Tree

onnxbot-worker-3 mentioned this pull request Feb 1, 2018

[auto] pytorch-pr-4999 onnxbot/onnx-fb-universe#517

Closed

Replace async with non_blocking for Python 3.7 upgrade

37adccd

goldsborough force-pushed the async branch from 3550789 to 37adccd Compare February 1, 2018 22:17

goldsborough changed the title ~~Replace async with non_blocking for Python 3.7 upgrade~~ Replace async with non_blocking for Python 3.7 Feb 1, 2018

Remove trailing whitespace

532e9a1

Give _cuda and _type kwargs and accept async for compatibility

8ae9d54

Rename async to non_blocking in all C++ code

a78748f

colesbury reviewed Feb 2, 2018

View reviewed changes

Add entries for async in python_variable_methods

abc41c0

ezyang reviewed Feb 2, 2018

View reviewed changes

torch/_utils.py Outdated

"""

if kwargs:

if len(kwargs) != 1 or 'async' not in kwargs:

raise ValueError("kwargs may only contain the key 'async'")

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

ezyang approved these changes Feb 2, 2018

View reviewed changes

Friendlier backward compatibility for cuda and type

579275a

soumith merged commit 86fd5fd into pytorch:master Feb 2, 2018

soumith mentioned this pull request Feb 11, 2018

async keyword on cuda() no longer works #5174

Closed

cclauss pushed a commit to cclauss/pytorch that referenced this pull request Jul 27, 2018

Travis CI: Run flake on Python 2.7 and 3.7

e9afc1a

Flake8 will produce different results on Python 2 and 3. Python 3.7 has __async__ as a reserved word pytorch#4999.

This was referenced Jul 27, 2018

Travis CI: Run flake on Python 2.7 and 3.7 #9953

Closed

print() is a function in Python 3 albertpumarola/GANimation#1

Closed

cclauss mentioned this pull request Jul 27, 2018

Replace async with non_blocking for Python 3.7 quark0/darts#25

Closed

cclauss pushed a commit to cclauss/IBN-Net that referenced this pull request Jul 28, 2018

cuda(async=True) --> cuda(non_blocking=True)

c33c13f

XingangPan#2 and pytorch/pytorch#4999

cclauss mentioned this pull request Jul 28, 2018

cuda(async=True) --> cuda(non_blocking=True) XingangPan/IBN-Net#3

Closed

cclauss mentioned this pull request Sep 25, 2018

'async' is a reserved word in Python 3.7 catalyst-team/catalyst#3

Closed

cclauss pushed a commit to cclauss/prometheus that referenced this pull request Sep 28, 2018

x.cuda(async=True) —> x.cuda(non_blocking=True)

277ba2d

Fixes catalyst-team#3 The Dockerfiles are pip installing PyTorch=0.41 so this change is appropriate to match with pytorch/pytorch#4999

cclauss mentioned this pull request Sep 28, 2018

x.cuda(async=True) —> x.cuda(non_blocking=True) catalyst-team/catalyst#4

Closed

cclauss pushed a commit to cclauss/pytorch that referenced this pull request Sep 28, 2018

Docs: Change cuda(async) —> cuda(non_blocking)

fc02661

Modify the docs to match the changes made in pytorch#4999

This was referenced Sep 28, 2018

Docs: Change cuda(async) —> cuda(non_blocking) #12158

Closed

Add support for Python 3.7 on PyTourch 0.4.1 andreasveit/convnet-aig#2

Closed

cclauss mentioned this pull request Nov 28, 2018

'async' is a reserved word in Python >= 3.7 CSAILVision/IBD#1

Closed

cclauss mentioned this pull request Nov 30, 2018

cuda(async) —> cuda(non_blocking) for Python >= 3.7 CSAILVision/semantic-segmentation-pytorch#129

Merged

This was referenced Jan 9, 2019

cuda(async) —> cuda(non_blocking) for Python >= 3.7 rowanz/r2c#2

Merged

cuda(async) —> cuda(non_blocking) for Python >= 3.7 microsoft/samples-for-ai#152

Merged

cclauss mentioned this pull request Mar 6, 2019

python3.7 support Miserlou/Zappa#1762

Merged

cclauss mentioned this pull request May 4, 2019

'async' is a reserved word in Python >= 3.7 ialhashim/DenseDepth#17

Merged

cclauss mentioned this pull request Jun 16, 2019

'async' is a reserved word in Python >= 3.7 zuoym15/craves.ai#2

Merged

cclauss mentioned this pull request Jun 18, 2019

'async' is a reserved word in Python >= 3.7 microsoftarchive/BatchAI#66

Closed

cclauss added a commit to cclauss/OmniNet that referenced this pull request Jul 24, 2019

Syntax Error: "async" is a keyword in Python >= 3.7

4afb0e9

Using __async__ in this way is a Syntax Error in Python >= 3.7. This change tracks with the changes made in pytorch/pytorch#4999

cclauss mentioned this pull request Jul 24, 2019

Syntax Error: "async" is a keyword in Python >= 3.7 subho406/OmniNet#1

Merged

cclauss mentioned this pull request Aug 19, 2019

'async' is a reserved word in Python >= 3.7 LiyuanLucasLiu/RAdam#13

Merged

cclauss mentioned this pull request Nov 29, 2019

'async' is a reserved word in Python >= 3.7 SeuTao/RSNA2019_Intracranial-Hemorrhage-Detection#2

Closed

Blade6570 mentioned this pull request Sep 10, 2020

Syntax Error - pix2pix_model.py Blade6570/icface#15

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Replace async with non_blocking for Python 3.7#4999

Replace async with non_blocking for Python 3.7#4999
soumith merged 6 commits intopytorch:masterfrom
goldsborough:async

goldsborough commented Feb 1, 2018 •

edited

Loading

Uh oh!

ezyang commented Feb 1, 2018 •

edited

Loading

Uh oh!

ezyang commented Feb 1, 2018

Uh oh!

goldsborough commented Feb 1, 2018

Uh oh!

goldsborough commented Feb 1, 2018

Uh oh!

goldsborough commented Feb 2, 2018

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

ezyang left a comment

Uh oh!

goldsborough commented Feb 2, 2018 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

goldsborough commented Feb 1, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ezyang commented Feb 1, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ezyang commented Feb 1, 2018

Uh oh!

goldsborough commented Feb 1, 2018

Uh oh!

goldsborough commented Feb 1, 2018

Uh oh!

goldsborough commented Feb 2, 2018

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

ezyang left a comment

Choose a reason for hiding this comment

Uh oh!

goldsborough commented Feb 2, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

goldsborough commented Feb 1, 2018 •

edited

Loading

ezyang commented Feb 1, 2018 •

edited

Loading

goldsborough commented Feb 2, 2018 •

edited

Loading