Add complex hooks for out of tree complex implementation. by ezyang · Pull Request #11216 · pytorch/pytorch

ezyang · 2018-09-04T14:25:53Z

This PR adds a hooks interface for registering types for complex
scalar types, and a sample implementation of the hook in
test_cpp_extensions.

The hook registration is patterned off of the existing CUDA hooks.

Signed-off-by: Edward Z. Yang ezyang@fb.com

CC @Roger-luo

ssnl

We discussed offline. In future patch(s), we will

Support ATen/native the tensor factory functions for complex numbers. Need to port 5~10 functions, e.g., resize_.
Default register our own bare bones complex type. Test that we can swap type pointers in when loading the complex extension.

ezyang · 2018-09-04T15:35:40Z

@pytorchbot retest this please

ezyang · 2018-09-04T16:21:12Z

@pytorchbot retest this please

ssnl · 2018-09-04T17:00:50Z

errors look legit:


16:56:19 /tmp/torch_extensions/complex_registration_extension/main.cpp:29:64: error: expected class-name before '{' token
16:56:19          struct CPUComplexFloatType : public at::CPUTypeDefault {
16:56:19                                                                 ^
16:56:19 /tmp/torch_extensions/complex_registration_extension/main.cpp:34:22: error: 'at::ScalarType at::CPUComplexFloatType::scalarType() const' marked override, but does not override
16:56:19            ScalarType scalarType() const override;
16:56:19                       ^
16:56:19 /tmp/torch_extensions/complex_registration_extension/main.cpp:35:19: error: 'at::Backend at::CPUComplexFloatType::backend() const' marked override, but does not override
16:56:19            Backend backend() const override;
16:56:19                    ^
16:56:19 /tmp/torch_extensions/complex_registration_extension/main.cpp:36:24: error: 'const char* at::CPUComplexFloatType::toString() const' marked override, but does not override
16:56:19            const char * toString() const override;
16:56:19                         ^
16:56:19 /tmp/torch_extensions/complex_registration_extension/main.cpp:37:18: error: 'size_t at::CPUComplexFloatType::elementSizeInBytes() const' marked override, but does not override
16:56:19            size_t elementSizeInBytes() const override;
...

gchanan · 2018-09-04T17:23:16Z

does it make sense to call it data_type instead of scalar_t? That's the c10 concept right?

ezyang · 2018-09-04T20:20:08Z

@gchanan Yeah. We should probably do a global search replaced of scalar_t to data_t.

Roger-luo · 2018-09-04T20:26:14Z

@ezyang I had an old issue for renaming real and a python script for auto-renaming that might be helpful.

#4575

ezyang · 2018-09-04T20:57:35Z

@Roger-luo The fix is landed on master :)

This PR adds a hooks interface for registering types for complex scalar types, and a sample implementation of the hook in test_cpp_extensions. The hook registration is patterned off of the existing CUDA hooks. Signed-off-by: Edward Z. Yang <ezyang@fb.com>

Signed-off-by: Edward Z. Yang <ezyang@fb.com>

facebook-github-bot

ezyang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

ezyang · 2018-09-05T14:17:31Z

@pytorchbot retest this please

Signed-off-by: Edward Z. Yang <ezyang@fb.com>

facebook-github-bot

ezyang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

Summary: This PR adds a hooks interface for registering types for complex scalar types, and a sample implementation of the hook in test_cpp_extensions. The hook registration is patterned off of the existing CUDA hooks. Signed-off-by: Edward Z. Yang <ezyang@fb.com> CC The controller you requested could not be found. Pull Request resolved: pytorch/pytorch#11216 Differential Revision: D9654840 Pulled By: ezyang fbshipit-source-id: 7b97646280d584f8ed6e14ee10a4abcd04cf2987

resolve conflict in data parallel model * master: (201 commits) Add cost inference to ConvGradient and WeightedSum operators (pytorch#10744) Move collapse dims into a single place (pytorch#11272) Fix some more warnings (pytorch#11257) Fix the batchnorm onnx exporting when affine=False Improve error message to include return types too (pytorch#11245) Check doxygen output in travis (pytorch#11124) Accept more numpy scalars as doubles (pytorch#9659) Fixed log message (pytorch#10874) Fix to distribution.__repr__ with lazy attributes (pytorch#11263) Add import export step to end to end tests Add complex hooks for out of tree complex implementation. (pytorch#11216) Unify opt flag for cmake codegen (pytorch#11227) nomnigraph - fix memory error in NN subgraph matchOp (pytorch#11127) Port PackedSequences functions to C++ (pytorch#11224) Treat numerical differences as warnings instead of errors when tracing (pytorch#11246) add a Float16UniformFill (pytorch#11123) Implement torch.tensordot (pytorch#10025) keep net type info when generating model complete net (pytorch#11032) Get rid of some uses of type() (pytorch#11215) Reorganize methods in Type, add CPUTypeDefault/CUDATypeDefault (pytorch#11205) ...

) Summary: This PR adds a hooks interface for registering types for complex scalar types, and a sample implementation of the hook in test_cpp_extensions. The hook registration is patterned off of the existing CUDA hooks. Signed-off-by: Edward Z. Yang <ezyang@fb.com> CC The controller you requested could not be found. Pull Request resolved: pytorch#11216 Differential Revision: D9654840 Pulled By: ezyang fbshipit-source-id: 7b97646280d584f8ed6e14ee10a4abcd04cf2987

ezyang requested review from apaszke, colesbury, gchanan, soumith and zdevito as code owners September 4, 2018 14:25

Roger-luo mentioned this pull request Sep 4, 2018

Use native pytorch complex type PIQuIL/QuCumber#53

Closed

ssnl approved these changes Sep 4, 2018

View reviewed changes

gchanan approved these changes Sep 4, 2018

View reviewed changes

ezyang force-pushed the pr/complex-hooks branch from efb345f to a05ebdf Compare September 4, 2018 22:35

ezyang added 2 commits September 5, 2018 06:50

Fix lint

81d426d

Signed-off-by: Edward Z. Yang <ezyang@fb.com>

ezyang force-pushed the pr/complex-hooks branch from a05ebdf to 81d426d Compare September 5, 2018 13:59

facebook-github-bot reviewed Sep 5, 2018

View reviewed changes

fix forward declare

96df303

Signed-off-by: Edward Z. Yang <ezyang@fb.com>

facebook-github-bot reviewed Sep 5, 2018

View reviewed changes

facebook-github-bot closed this in 1808e36 Sep 5, 2018

ezyang added the merged label Jun 26, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add complex hooks for out of tree complex implementation.#11216

Add complex hooks for out of tree complex implementation.#11216
ezyang wants to merge 3 commits intopytorch:masterfrom
ezyang:pr/complex-hooks

ezyang commented Sep 4, 2018

Uh oh!

ssnl left a comment

Uh oh!

ezyang commented Sep 4, 2018

Uh oh!

ezyang commented Sep 4, 2018

Uh oh!

ssnl commented Sep 4, 2018

Uh oh!

gchanan commented Sep 4, 2018

Uh oh!

ezyang commented Sep 4, 2018

Uh oh!

Roger-luo commented Sep 4, 2018

Uh oh!

ezyang commented Sep 4, 2018

Uh oh!

facebook-github-bot left a comment

Uh oh!

ezyang commented Sep 5, 2018

Uh oh!

facebook-github-bot left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

ezyang commented Sep 4, 2018

Uh oh!

ssnl left a comment

Choose a reason for hiding this comment

Uh oh!

ezyang commented Sep 4, 2018

Uh oh!

ezyang commented Sep 4, 2018

Uh oh!

ssnl commented Sep 4, 2018

Uh oh!

gchanan commented Sep 4, 2018

Uh oh!

ezyang commented Sep 4, 2018

Uh oh!

Roger-luo commented Sep 4, 2018

Uh oh!

ezyang commented Sep 4, 2018

Uh oh!

facebook-github-bot left a comment

Choose a reason for hiding this comment

Uh oh!

ezyang commented Sep 5, 2018

Uh oh!

facebook-github-bot left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants