[complex] conv1d #75013

kshitij12345 · 2022-03-31T12:05:27Z

Reference: #71108

facebook-github-bot · 2022-03-31T12:05:33Z

🔗 Helpful links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/75013
📄 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

💊 CI failures summary and remediations

As of commit e8cb463 (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).

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

Click here to manually regenerate this comment.

kshitij12345 · 2022-04-01T11:22:49Z

torch/testing/_internal/common_methods_invocations.py

@@ -11191,6 +11192,9 @@ def ref_pairwise_distance(input1, input2):
               # RuntimeError: !lhs.isAliasOf(rhs)INTERNAL ASSERT FAILED at
               # "../torch/csrc/jit/passes/utils/check_alias_annotation.cpp":103, please report a bug to PyTorch.
               DecorateInfo(unittest.skip("Skipped!"), 'TestJit', 'test_variant_consistency_jit'),
+               # AssertionError: None mismatch: torch.complex128 is not None
+               DecorateInfo(unittest.expectedFailure, 'TestDtypeCustomRules',
+                            'test_custom_rules', dtypes=(torch.complex64, torch.complex128)),


Complete Stack Trace:

raceback (most recent call last): File "/home/kshiteej/Pytorch/pytorch_complex_convolution.py/test/jit/test_dtype_analysis.py", line 361, in test_custom_rules self.custom_rules_test_base(device, dtype, op) File "/home/kshiteej/Pytorch/pytorch_complex_convolution.py/test/jit/test_dtype_analysis.py", line 357, in custom_rules_test_base self.assert_output_dtype_equal(expected_res, graph) File "/home/kshiteej/Pytorch/pytorch_complex_convolution.py/test/jit/test_dtype_analysis.py", line 323, in assert_output_dtype_equal self.assert_tensor_dtype_equal(expected_res, actual_dtype[0]) File "/home/kshiteej/Pytorch/pytorch_complex_convolution.py/test/jit/test_dtype_analysis.py", line 332, in assert_tensor_dtype_equal self.assertEqual(tensor_output.dtype, graph_dtype) File "/home/kshiteej/.conda/envs/pytorch-cuda-dev/lib/python3.10/site-packages/torch/testing/_internal/common_utils.py", line 2160, in assertEqual assert_equal( File "/home/kshiteej/.conda/envs/pytorch-cuda-dev/lib/python3.10/site-packages/torch/testing/_comparison.py", line 1084, in assert_equal raise error_metas[0].to_error() AssertionError: None mismatch: torch.complex128 is not None

what does TestDtypeCustomRules. test_custom_rules test? why is this an expected failure?

I am not sure. Looking at the test it seems to check if dtype is propagated correctly in JIT traced graph. (I was planning to point someone from the JIT team to unblock this PR)

pytorch/test/jit/test_dtype_analysis.py

Lines 334 to 357 in ad028e5

def custom_rules_test_base(self, device, dtype, op, allow_eager_fail=False):

try:

samples = op.sample_inputs(device, dtype, requires_grad=False)

sample_input = first_sample(self, samples)

input_args = [sample_input.input, *sample_input.args]

expected_res = op(*input_args, **sample_input.kwargs)

except Exception as e:

if allow_eager_fail:

return

else:

raise e

func = op.get_op()

traced_fn = create_traced_fn(self, func)

# Have to run the traced function to actually generate the trace

traced_fn(sample_input.input, *sample_input.args, **sample_input.kwargs)

# Run the Dtype Analysis

graph = traced_fn.graph # Note this is a cached graph

input_tensors = [t for t in input_args if isinstance(t, torch.Tensor)]

self.prop_dtype_on_graph(graph, input_tensors)

self.assert_output_dtype_equal(expected_res, graph)

kshitij12345 · 2022-04-01T15:23:42Z

@anjali411 can you please take a look to see if it looks in the right direction. Thanks!

aten/src/ATen/native/Convolution.cpp

anjali411 · 2022-04-03T00:44:35Z

aten/src/ATen/native/Convolution.cpp

-  auto output = at::_convolution_mode(
-      input, weight, bias, stride, std::move(padding), dilation, groups);
+  Tensor output;
+  if (at::isComplexType(input_.scalar_type())) {


nice so adds support for complex half too

Not at this point as +, -, *, view_as_real will need to be supported first.

anjali411 · 2022-04-03T00:45:00Z

torch/testing/_internal/common_methods_invocations.py

-           dtypes=floating_types_and(torch.int64),
-           dtypesIfCUDA=floating_types_and(torch.float16, *[torch.bfloat16] if (CUDA11OrLater or TEST_WITH_ROCM) else []),
+           dtypes=floating_and_complex_types_and(torch.int64),
+           dtypesIfCUDA=floating_and_complex_types_and(torch.float16,


we should also test complex half

anjali411

Granted that this would be slow (since we call into three convolution kernels), however it's a cool functionality and has been asked by several users so pretty exciting!

This PR looks good to me but we also need more tests to validate all cases (extend tests in test_nn.py) as well as a test to verify if we are computing the correct value.

kshitij12345 · 2022-04-05T14:12:08Z

@anjali411 have updated the relevant tests in test_nn. Have also added a test for correctness against scipy. PTAL :)

0x00b1 · 2022-04-05T15:49:59Z

Very excited! Nice job @kshitij12345 !

anjali411 · 2022-04-05T17:25:54Z

test/test_nn.py

+
+        # Global dtype for this test suite is torch.double
+        # This leads to change in type-promotion
+        # and conv1d outputs `complex128` for `complex64` input.


anjali411

great work @kshitij12345

anjali411 · 2022-04-05T17:28:33Z

@pytorchbot merge this please

github-actions · 2022-04-05T17:30:42Z

Hey @kshitij12345.
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.

mruberry · 2022-04-05T20:17:21Z

@pytorchbot revert this please

This reverts commit b64e7de. Reverted #75013 on behalf of https://github.com/mruberry

Reland : #75013 Reference: #71108 Pull Request resolved: #75310 Approved by: https://github.com/anjali411

Summary: Reland : #75013 Reference: #71108 Pull Request resolved: #75310 Approved by: https://github.com/anjali411 Test Plan: contbuild & OSS CI, see https://hud.pytorch.org/commit/pytorch/pytorch/706b9e8b8d8b17cc33f74636bf520a5a53e4d386 Reviewed By: b0noI Differential Revision: D35437863 fbshipit-source-id: 068ca2191e3be2abac4082bae3b234b21eb1ac0d

[complex] conv1d

9a52316

facebook-github-bot added the cla signed label Mar 31, 2022

pytorchbot added the open source label Mar 31, 2022

kshitij12345 added 7 commits April 1, 2022 07:44

add complex path for str mode

633fa59

update OpInfo entry

98d710b

small refactor

fa4de3d

refactor

df874a9

make flake8 happy

a137e4f

retrigger CI

ebda8ba

skip failing jit tests

fc8d3a6

kshitij12345 commented Apr 1, 2022

View reviewed changes

anjali411 reviewed Apr 1, 2022

View reviewed changes

aten/src/ATen/native/Convolution.cpp Show resolved Hide resolved

anjali411 reviewed Apr 3, 2022

View reviewed changes

kshitij12345 added 4 commits April 4, 2022 07:44

update few existing tests to run for complex

e2ac66c

Merge branch 'pytorch:master' into develop/complex/conv1d

7f525aa

add correctness test

0525045

Merge branch 'develop/complex/conv1d' into develop/complex/conv1d

e8cb463

kshitij12345 marked this pull request as ready for review April 5, 2022 16:14

anjali411 reviewed Apr 5, 2022

View reviewed changes

anjali411 approved these changes Apr 5, 2022

View reviewed changes

pytorchmergebot closed this in b64e7de Apr 5, 2022

pytorchmergebot added the Reverted label Apr 5, 2022

pytorchmergebot added a commit that referenced this pull request Apr 5, 2022

Revert "[complex] conv1d"

862f674

This reverts commit b64e7de. Reverted #75013 on behalf of https://github.com/mruberry

kshitij12345 mentioned this pull request Apr 6, 2022

[reland] [complex] conv1d #75310

Closed

pytorchmergebot pushed a commit that referenced this pull request Apr 6, 2022

[reland] [complex] conv1d

706b9e8

Reland : #75013 Reference: #71108 Pull Request resolved: #75310 Approved by: https://github.com/anjali411

	def custom_rules_test_base(self, device, dtype, op, allow_eager_fail=False):
	try:
	samples = op.sample_inputs(device, dtype, requires_grad=False)
	sample_input = first_sample(self, samples)
	input_args = [sample_input.input, *sample_input.args]
	expected_res = op(input_args, *sample_input.kwargs)

	except Exception as e:
	if allow_eager_fail:
	return
	else:
	raise e

	func = op.get_op()
	traced_fn = create_traced_fn(self, func)

	# Have to run the traced function to actually generate the trace
	traced_fn(sample_input.input, sample_input.args, *sample_input.kwargs)

	# Run the Dtype Analysis
	graph = traced_fn.graph # Note this is a cached graph
	input_tensors = [t for t in input_args if isinstance(t, torch.Tensor)]
	self.prop_dtype_on_graph(graph, input_tensors)
	self.assert_output_dtype_equal(expected_res, graph)

[complex] conv1d #75013

[complex] conv1d #75013

Uh oh!

Conversation

kshitij12345 commented Mar 31, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

facebook-github-bot commented Mar 31, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful links

💊 CI failures summary and remediations

Uh oh!

kshitij12345 Apr 1, 2022

Choose a reason for hiding this comment

Uh oh!

anjali411 Apr 3, 2022

Choose a reason for hiding this comment

Uh oh!

kshitij12345 Apr 3, 2022

Choose a reason for hiding this comment

Uh oh!

kshitij12345 commented Apr 1, 2022

Uh oh!

Uh oh!

anjali411 Apr 3, 2022

Choose a reason for hiding this comment

Uh oh!

kshitij12345 Apr 3, 2022

Choose a reason for hiding this comment

Uh oh!

anjali411 Apr 3, 2022

Choose a reason for hiding this comment

Uh oh!

anjali411 left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kshitij12345 commented Apr 5, 2022

Uh oh!

0x00b1 commented Apr 5, 2022

Uh oh!

anjali411 Apr 5, 2022

Choose a reason for hiding this comment

Uh oh!

anjali411 left a comment

Choose a reason for hiding this comment

Uh oh!

anjali411 commented Apr 5, 2022

Uh oh!

github-actions bot commented Apr 5, 2022

Uh oh!

mruberry commented Apr 5, 2022

Uh oh!

Uh oh!

kshitij12345 commented Mar 31, 2022 •

edited

Loading

facebook-github-bot commented Mar 31, 2022 •

edited

Loading

anjali411 left a comment •

edited

Loading