torch.fft: Two dimensional FFT functions #45164

peterbell10 · 2020-09-22T21:19:56Z

This PR implements fft2, ifft2, rfft2 and irfft2. These are the last functions required for torch.fft to match numpy.fft. If you look at either NumPy or SciPy you'll see that the 2-dimensional variants are identical to *fftn in every way, except for the default value of axes. In fact you can even use fft2 to do general n-dimensional transforms.

Stack from ghstack:

torch.fft: Two dimensional FFT functions #45164 torch.fft: Two dimensional FFT functions

Differential Revision: D24363639

[ghstack-poisoned]

ghstack-source-id: 0fc505d Pull Request resolved: #45164

aten/src/ATen/native/native_functions.yaml

dr-ci · 2020-09-22T23:49:43Z

💊 CI failures summary and remediations

As of commit bc3bd1b (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.

See how this bot performed.

This comment has been revised 47 times.

This PR implements `fft2`, `ifft2`, `rfft2` and `irfft2`. These are the last functions required for `torch.fft` to match `numpy.fft`. If you look at either NumPy or SciPy you'll see that the 2-dimensional variants are identical to `*fftn` in every way, except for the default value of `axes`. In fact you can even use `fft2` to do general n-dimensional transforms. [ghstack-poisoned]

ghstack-source-id: 65f0767 Pull Request resolved: #45164

This PR implements `fft2`, `ifft2`, `rfft2` and `irfft2`. These are the last functions required for `torch.fft` to match `numpy.fft`. If you look at either NumPy or SciPy you'll see that the 2-dimensional variants are identical to `*fftn` in every way, except for the default value of `axes`. In fact you can even use `fft2` to do general n-dimensional transforms. [ghstack-poisoned]

ghstack-source-id: 923c39c Pull Request resolved: #45164

This PR implements `fft2`, `ifft2`, `rfft2` and `irfft2`. These are the last functions required for `torch.fft` to match `numpy.fft`. If you look at either NumPy or SciPy you'll see that the 2-dimensional variants are identical to `*fftn` in every way, except for the default value of `axes`. In fact you can even use `fft2` to do general n-dimensional transforms. [ghstack-poisoned]

ghstack-source-id: 60b5fec Pull Request resolved: #45164

This PR implements `fft2`, `ifft2`, `rfft2` and `irfft2`. These are the last functions required for `torch.fft` to match `numpy.fft`. If you look at either NumPy or SciPy you'll see that the 2-dimensional variants are identical to `*fftn` in every way, except for the default value of `axes`. In fact you can even use `fft2` to do general n-dimensional transforms. [ghstack-poisoned]

ghstack-source-id: 384b212 Pull Request resolved: #45164

mruberry · 2020-10-12T16:47:24Z

test/test_spectral_ops.py

+    @skipCPUIfNoMkl
+    @skipCUDAIfRocm
+    @onlyOnCPUAndCUDA
+    @unittest.skipIf(not TEST_NUMPY, 'NumPy not found')


We now require NumPy in our test suite and don't need these checks anymore

mruberry · 2020-10-12T16:51:21Z

test/test_spectral_ops.py

+
+                # Once with dim defaulted
+                input_np = input.cpu().numpy()
+                expected = numpy_fn(input_np, s, axes=(-2, -1), norm=norm)


This is interesting. Should this not explicitly provide a value for axes to be consistent with the actual computation below?

mruberry · 2020-10-12T16:52:56Z

torch/csrc/api/include/torch/fft.h

  return torch::fft_ifft(self, n, dim, norm);
 }

+/// Computes the 2 dimensional fast Fourier transform over given dimensions.


"over given" -> "over the given"

"2 dimensional" -> "2-dimensional"

mruberry · 2020-10-12T16:53:09Z

torch/csrc/api/include/torch/fft.h

+  return torch::fft_fft2(self, s, dim, norm);
+}
+
+/// Computes the 2 dimensional fast Fourier transform over given dimensions.


Computes the inverse of torch.fft.fft2.

mruberry · 2020-10-12T16:54:12Z

torch/csrc/api/include/torch/fft.h

  return torch::fft_irfft(self, n, dim, norm);
 }

+/// Computes the 2 dimensional FFT of real input with onesided Hermitian output.


Should this be: "of a real input. Returns a onesided Hermitian output."?

mruberry · 2020-10-12T16:55:18Z

torch/csrc/api/include/torch/fft.h

+/// Example:
+/// ```
+/// auto t = torch::randn({128, 128}, dtype=kComplexDouble);
+/// torch::fft::irfftn(t);


This should be a call to irfft2, not irfftn

mruberry · 2020-10-12T16:56:12Z

torch/fft/__init__.py

+fft2 = _add_docstr(_fft.fft_fft2, r"""
+fft2(input, s=None, dim=(-2, -1), norm=None) -> Tensor
+
+Computes the 2 dimensional discrete Fourier transform of :attr:`input`.


"2 dimensional" -> "2-dimensional"

mruberry · 2020-10-12T16:57:33Z

torch/fft/__init__.py

+Computes the 2 dimensional discrete Fourier transform of :attr:`input`.
+
+Note:
+    :func:`~torch.fft.fft2` is a convenience alias for :func:`~torch.fft.fftn`


Instead of making this a note, how about adding a second sentence to the preamble like: "Equivalent to ..." and linking fftn and showing how to map the arguments from this to fftn?

@mruberry PTAL, I've changed it but am not entirely sure about "map[ping] the arguments from this to fftn". The arguments should be exactly the same, other than dim not accepting None.

torch/fft/__init__.py

mruberry · 2020-10-12T16:59:41Z

Hey @peterbell10!

Great to have these last few functions. As usual I've asked a few questions about the docs. Sorry about the review delay but review responsiveness should finally be back to normal. Looking forward to hearing your thoughts!

This PR implements `fft2`, `ifft2`, `rfft2` and `irfft2`. These are the last functions required for `torch.fft` to match `numpy.fft`. If you look at either NumPy or SciPy you'll see that the 2-dimensional variants are identical to `*fftn` in every way, except for the default value of `axes`. In fact you can even use `fft2` to do general n-dimensional transforms. [ghstack-poisoned]

ghstack-source-id: e8c06f7 Pull Request resolved: #45164

mruberry · 2020-10-16T04:35:38Z

torch/fft/__init__.py

+""")
+
+ifft2 = _add_docstr(_fft.fft_ifft2, r"""
+ifftn(input, s=None, dim=(-2, -1), norm=None) -> Tensor


ifftn->ifft2

mruberry

Another awesome PR, @peterbell10! Just one minor correction in the docs. Ping me when you're ready.

This PR implements `fft2`, `ifft2`, `rfft2` and `irfft2`. These are the last functions required for `torch.fft` to match `numpy.fft`. If you look at either NumPy or SciPy you'll see that the 2-dimensional variants are identical to `*fftn` in every way, except for the default value of `axes`. In fact you can even use `fft2` to do general n-dimensional transforms. [ghstack-poisoned]

peterbell10 · 2020-10-16T11:42:40Z

Fixed. Thanks @mruberry.

This PR implements `fft2`, `ifft2`, `rfft2` and `irfft2`. These are the last functions required for `torch.fft` to match `numpy.fft`. If you look at either NumPy or SciPy you'll see that the 2-dimensional variants are identical to `*fftn` in every way, except for the default value of `axes`. In fact you can even use `fft2` to do general n-dimensional transforms. [ghstack-poisoned]

ghstack-source-id: 36b28d2 Pull Request resolved: #45164

facebook-github-bot · 2020-10-18T00:09:18Z

@mruberry merged this pull request in da95eec.

torch.fft: Two dimensional FFT functions

d8f44d0

[ghstack-poisoned]

peterbell10 requested review from ebetica, goldsborough and yf225 as code owners September 22, 2020 21:19

This was referenced Sep 22, 2020

torch.fft: Multi-dimensional transforms #44550

Closed

Deprecate old fft functions #44876

Closed

torch.fft helper functions #44877

Closed

codegen: Improve array default handing #45163

Closed

peterbell10 added a commit that referenced this pull request Sep 22, 2020

torch.fft: Two dimensional FFT functions

a23dc0a

ghstack-source-id: 0fc505d Pull Request resolved: #45164

peterbell10 marked this pull request as draft September 22, 2020 21:20

pytorchbot added the open source label Sep 22, 2020

peterbell10 added the fft label Sep 22, 2020

peterbell10 commented Sep 22, 2020

View reviewed changes

aten/src/ATen/native/native_functions.yaml Outdated Show resolved Hide resolved

peterbell10 requested a review from mruberry September 22, 2020 21:47

mruberry removed request for ebetica, goldsborough and yf225 September 22, 2020 22:20

peterbell10 added a commit that referenced this pull request Sep 23, 2020

torch.fft: Two dimensional FFT functions

37e4652

ghstack-source-id: 65f0767 Pull Request resolved: #45164

peterbell10 added a commit that referenced this pull request Sep 23, 2020

torch.fft: Two dimensional FFT functions

e64451c

ghstack-source-id: 923c39c Pull Request resolved: #45164

peterbell10 added a commit that referenced this pull request Oct 5, 2020

torch.fft: Two dimensional FFT functions

39070d9

ghstack-source-id: 60b5fec Pull Request resolved: #45164

mruberry added module: fft and removed fft labels Oct 6, 2020

peterbell10 added a commit that referenced this pull request Oct 7, 2020

torch.fft: Two dimensional FFT functions

29a119c

ghstack-source-id: 384b212 Pull Request resolved: #45164

mruberry reviewed Oct 12, 2020

View reviewed changes

torch/fft/__init__.py Show resolved Hide resolved

peterbell10 added a commit that referenced this pull request Oct 12, 2020

torch.fft: Two dimensional FFT functions

850730e

ghstack-source-id: e8c06f7 Pull Request resolved: #45164

mruberry reviewed Oct 16, 2020

View reviewed changes

mruberry approved these changes Oct 16, 2020

View reviewed changes

peterbell10 added a commit that referenced this pull request Oct 16, 2020

torch.fft: Two dimensional FFT functions

f3a26ee

ghstack-source-id: 36b28d2 Pull Request resolved: #45164

facebook-github-bot closed this in da95eec Oct 17, 2020

facebook-github-bot added the Merged label Oct 18, 2020

facebook-github-bot deleted the gh/peterbell10/14/head branch October 21, 2020 14:20

torch.fft: Two dimensional FFT functions #45164

torch.fft: Two dimensional FFT functions #45164

Uh oh!

Conversation

peterbell10 commented Sep 22, 2020 • edited by mruberry Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

dr-ci bot commented Sep 22, 2020 • edited by facebook-github-bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

💊 CI failures summary and remediations

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mruberry Oct 12, 2020 • 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

mruberry commented Oct 12, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mruberry left a comment

Choose a reason for hiding this comment

Uh oh!

peterbell10 commented Oct 16, 2020

Uh oh!

facebook-github-bot commented Oct 18, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

peterbell10 commented Sep 22, 2020 •

edited by mruberry

Loading

dr-ci bot commented Sep 22, 2020 •

edited by facebook-github-bot

Loading

mruberry Oct 12, 2020 •

edited

Loading