platform/upstream/pytorch.git
3 years agoMove fx2trt and oss_acc_tracer to oss (#63101)
Shiyan Deng [Sun, 15 Aug 2021 18:52:20 +0000 (11:52 -0700)]
Move fx2trt and oss_acc_tracer to oss (#63101)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63101

Move internal fx2trt to torch/fx/experimental/fx2trt and merge the two TRT interpreter we have right now. cc: mortzur as this might affect uru exporting script.

Move oss_acc_tracer to torch/fx/experimental/fx_acc.

Test Plan: CI

Reviewed By: jerryzh168

Differential Revision: D30257909

fbshipit-source-id: 4e374965fbf88d72e91844d9e9b6ff9b98f467d1

3 years agoHide all symbols in llvm namespace (#63272)
Bert Maher [Sun, 15 Aug 2021 18:28:23 +0000 (11:28 -0700)]
Hide all symbols in llvm namespace (#63272)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/63272

Test Plan: Imported from OSS

Reviewed By: nikithamalgifb

Differential Revision: D30331695

Pulled By: bertmaher

fbshipit-source-id: d35130c96f7e2a31fa86d9d80de59002e96301df

3 years agoAdd copy button to code snippets in docs (#63149)
anjali411 [Sun, 15 Aug 2021 13:22:53 +0000 (06:22 -0700)]
Add copy button to code snippets in docs (#63149)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/63149

Test Plan: Imported from OSS

Reviewed By: navahgar, albanD

Differential Revision: D30308891

Pulled By: anjali411

fbshipit-source-id: ad51180ab2f27c4525682b2603bbf753bb8f1ce9

3 years ago[Pytorch Edge] Enable kineto profiler on mobile via EdgeKinetoProfiler (#62419)
Kimish Patel [Sat, 14 Aug 2021 04:37:57 +0000 (21:37 -0700)]
[Pytorch Edge] Enable kineto profiler on mobile via EdgeKinetoProfiler (#62419)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62419

This diff adds support for cpu only kineto profiler on mobile. Thus
enabling chrome trace generation on mobile. This bring cpp API for
mobile profiling on part with Torchscript.
This is done via:
1. Utilizating debug handle annotations in KinetoEvent.
2. Adding post processing capability, via callbacks, to
KinetoThreadLocalState
3. Creating new RAII stype profiler, KinetoEdgeCPUProfiler, which can be
used in surrounding scope of model execution. This will write chrome
trace to the location specified in profiler constructor.

Test Plan:
MobileProfiler.ModuleHierarchy

Imported from OSS

Reviewed By: raziel

Differential Revision: D29993660

fbshipit-source-id: 0b44f52f9e9c5f5aff81ebbd9273c254c3c03299

3 years ago[Pytorch Mobile] Combing instructions and debug hanles in single struct (#62418)
Kimish Patel [Sat, 14 Aug 2021 04:37:57 +0000 (21:37 -0700)]
[Pytorch Mobile] Combing instructions and debug hanles in single struct (#62418)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62418

Debug handles have one to one correspondence with instruction, so just
combine them in one.

Test Plan:
CI

Imported from OSS

Reviewed By: raziel

Differential Revision: D29993661

fbshipit-source-id: 125c7163174cf66624dd95f110fdc8208fea8a07

3 years ago[Pytorch Profiler] Introduce scopes to enableProfiler (#62417)
Kimish Patel [Sat, 14 Aug 2021 04:37:57 +0000 (21:37 -0700)]
[Pytorch Profiler] Introduce scopes to enableProfiler (#62417)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62417

This diff adds an option to make enableProfiler enable callbacks only
for certain RecordScopes.
Why?
Profiling has some overhead when we repeatedly execute callbacks for
alls copes. On mobile side when we often have small quantized models
this overhead can be large. We observed that by only profiling top level
op and skipping profiling of other atend ops called within we can limit
this overhead. For example, instead of profling at::conv2d -> at::convolution ->
at::convolution_ and further more if ops like transpose etc. are called,
skipping profiling of those. Of course this limits the visibility, but
at the least this way we get a choice.

Test Plan: Imported from OSS

Reviewed By: ilia-cher

Differential Revision: D29993659

fbshipit-source-id: 852d3ae7822f0d94dc6e507bd4019b60d488ef69

3 years ago[Pytorch Profiler] Add debug_handles to KinetoEvent (#62228)
Kimish Patel [Sat, 14 Aug 2021 04:37:57 +0000 (21:37 -0700)]
[Pytorch Profiler] Add debug_handles to KinetoEvent (#62228)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62228

This diff adds debug handles to events and provides a way to use
RECORD_FUNCTIONs that will pass debug_handles down to profiler, which
will record it in the events.

Why add debug_handles?
For pytorch mobile, with lite interpreter, we generate debug handles
that can be used for lazily symbolicate exception traces to model level
stack trace. Similar to the model level stack trace you get in
TorchScript models. The debug_handles also enable getting module
hierarchy for lite interpreter model, support for which was added to
KinetoProfiler in previous diffs.

Followup plan:
1. Enabled scope callbacks such that lite interpreter can use it to
profiler only top level ops.
2. Enable post processing callbacks that take KinetoEvents and populate
module hierarchy using debug handles.

This will let us use KinetoProfiler for lite interpter use cases on
mobile. Aim is to use RAII guard to similarly generate chrome trace for
mobile usecases as well, although only for top level ops.

Test Plan:
test_misc : RecordDebugHandles.Basic

Imported from OSS

Reviewed By: ilia-cher

Differential Revision: D29935899

fbshipit-source-id: 4f06dc411b6b5fe0ffaebdd26d3274c96f8f389b

3 years ago[Pytorch Profiler] Move start timestamp to end of start callback (#62191)
Kimish Patel [Sat, 14 Aug 2021 04:37:57 +0000 (21:37 -0700)]
[Pytorch Profiler] Move start timestamp to end of start callback (#62191)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62191

This moves start timestamping to end of callback. This way we dont
account for callstack/module hierarchy related overhead in op runtime.

Test Plan:
CI

Imported from OSS

Reviewed By: ilia-cher

Differential Revision: D29910519

fbshipit-source-id: f462031a81ae12b3db7993cf482e5ad93a35e096

3 years ago[Pytorch Profiler] Add support for adding module hierarchy to (#61792)
Kimish Patel [Sat, 14 Aug 2021 04:37:57 +0000 (21:37 -0700)]
[Pytorch Profiler] Add support for adding module hierarchy to (#61792)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/61792

KinetoEvent

This PR adds module hierarchy information to events.
What is module hierarchy information attached to events?
During profiling a TorchScript module, when events are added, we ask JIT
what is the module hierarchy associated with the node being
executed. At the time of execution of that node, there might be multiple
frames in the stack of interpreter. For each frame, we find
corresponding node and the corresponding module hierarchy is queried.
Module hierarchy corresponding to the node is associated with node's
InlinedCallStack. InlinedCallStack of node tracks the path via which the
node is inlined. Thus during the inlining process we annotate
module information corresponding to the CallMethod nodes being inlined.

With this PR, chrome trace will contain additional metadata:
"Module Hierarchy". This can look like this:
TOP(ResNet)::forward.SELF(ResNet)::_forward_impl.layer1(Sequential)::forward.0(BasicBlock)::forward.conv1(Conv2d)::forward.SELF(Conv2d)::_conv_forward
It contains module instance, type name and the method name in the
callstack.

Test Plan:
test_profiler

Imported from OSS

Reviewed By: raziel, ilia-cher

Differential Revision: D29745442

fbshipit-source-id: dc8dfaf7c5b8ab256ff0b2ef1e5ec265ca366528

3 years agoadd substract of max and testcase (#63132)
leslie-fang-intel [Sat, 14 Aug 2021 03:49:27 +0000 (20:49 -0700)]
add substract of max and testcase (#63132)

Summary:
As discussed here https://github.com/pytorch/pytorch/pull/62897, in the path of BF16/non-last-dim Softmax, we miss the subtractions of max value which will cause the overflow in the `exp()` calculation when the value of input tensor is large, such as `1000.0`.
To avoid this issue, we add the subtractions of max value and the corresponding test cases in this PR.

Note w/o subtractions of max value(accidental reverts or changes), we will get the underlying error message of the test case
```
AssertionError: False is not true : Tensors failed to compare as equal!With rtol=0.05 and atol=0.05, found 103984 element(s) (out of 126720) whose difference(s) exceeded the margin of error (including 103984 nan comparisons). The greatest difference was nan (0.0 vs. nan), which occurred at index (0, 0, 0, 1).
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63132

Reviewed By: VitalyFedyunin

Differential Revision: D30280792

Pulled By: cpuhrsch

fbshipit-source-id: 722821debf983bbb4fec878975fa8a4da0d1d866

3 years agoOpInfo: `nn.functional.conv_transpose2d` (#62882)
Kushashwa Ravi Shrimali [Sat, 14 Aug 2021 00:10:07 +0000 (17:10 -0700)]
OpInfo: `nn.functional.conv_transpose2d` (#62882)

Summary:
See https://github.com/facebookresearch/functorch/issues/78 and https://github.com/pytorch/pytorch/issues/54261.

cc: mruberry zou3519 Chillee

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62882

Reviewed By: bdhirsh

Differential Revision: D30280804

Pulled By: zou3519

fbshipit-source-id: e40cdf43e98c1f11e45df6b8bc13110b4d29c45f

3 years agorefactor fx2trt example script so it can be imported as a library (#63262)
Kefei Lu [Fri, 13 Aug 2021 23:57:47 +0000 (16:57 -0700)]
refactor fx2trt example script so it can be imported as a library (#63262)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63262

Just create a `__main__` guard.

Test Plan: run linter, sandcastle tests

Reviewed By: 842974287

Differential Revision: D30263617

fbshipit-source-id: 8044ce5d815b043c3778591384cb13d9a89d0048

3 years ago[iOS] Add `LibTorch-Lite-Nightly` pod (#63239)
Hanton Yang [Fri, 13 Aug 2021 23:20:22 +0000 (16:20 -0700)]
[iOS] Add `LibTorch-Lite-Nightly` pod (#63239)

Summary:
D30090760 (https://github.com/pytorch/pytorch/commit/e182b459d94fe77c1d9f623c94fc2621c8cc55de) was reverted by D30303292 because of a lint issue in `LibTorch-Lite-Nightly.podspec.template`. Resubmit the diff after fixing the issue.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63239

Test Plan: Imported from OSS

Reviewed By: xta0

Differential Revision: D30315690

Pulled By: hanton

fbshipit-source-id: f0fa719ffc3b8181ab28c123584ae5c1da8992c0

3 years agoAllow TransformerEncoder and TransformerDecoder to accept 0-dim batch sized tensors...
Sameer Deshmukh [Fri, 13 Aug 2021 23:08:01 +0000 (16:08 -0700)]
Allow TransformerEncoder and TransformerDecoder to accept 0-dim batch sized tensors. (#62800)

Summary:
This issue fixes a part of https://github.com/pytorch/pytorch/issues/12013, which is summarized concretely in  https://github.com/pytorch/pytorch/issues/38115.

This PR allows TransformerEncoder and Decoder (alongwith the inner `Layer` classes) to accept inputs with 0-dimensional batch sizes.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62800

Reviewed By: VitalyFedyunin

Differential Revision: D30303240

Pulled By: jbschlosser

fbshipit-source-id: 8f8082a6f2a9f9d7ce0b22a942d286d5db62bd12

3 years ago[ROCm] Update HIP_VERSION to TORCH_HIP_VERSION (#62786)
Pruthvi Madugundu [Fri, 13 Aug 2021 21:57:17 +0000 (14:57 -0700)]
[ROCm] Update HIP_VERSION to TORCH_HIP_VERSION (#62786)

Summary:
- HIP_VERSION semantic versioning will change in ROCm4.3. The changes essentially remove the dependency on HIP_VERSION provided in the hip header to keep code compatible with older and newer versions of ROCm.
- TORCH_HIP_VERSION is derived from HIP_VERSION_MAJOR and HIP_VERSION_MINOR

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62786

Reviewed By: bdhirsh

Differential Revision: D30281682

Pulled By: seemethere

fbshipit-source-id: e41e69fb9e13de5ddd1af99ba5bbdcbb7b64b673

3 years agoRespect user-set CMAKE_PREFIX_PATH (#61904)
Can Balioglu [Fri, 13 Aug 2021 20:47:37 +0000 (13:47 -0700)]
Respect user-set CMAKE_PREFIX_PATH (#61904)

Summary:
Fixes the case where the `CMAKE_PREFIX_PATH` variable gets silently overwritten by a user specified environment variable.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/61904

Reviewed By: walterddr, malfet

Differential Revision: D29792014

Pulled By: cbalioglu

fbshipit-source-id: babacc8d5a1490bff1e14247850cc00c6ba9e6be

3 years agoRemove left-over print in test_diff_graph_inline_threshold (#63231)
gmagogsfm [Fri, 13 Aug 2021 20:06:08 +0000 (13:06 -0700)]
Remove left-over print in test_diff_graph_inline_threshold (#63231)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/63231

Reviewed By: VitalyFedyunin

Differential Revision: D30305851

Pulled By: gmagogsfm

fbshipit-source-id: 43da3b5f49ad4a6a2d6d174acf792f3ccf41a463

3 years agoAdd CostInferenceFunction for SplitOp (#63133)
Tanvir Zaman [Fri, 13 Aug 2021 19:25:16 +0000 (12:25 -0700)]
Add CostInferenceFunction for SplitOp (#63133)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63133

SplitOp is costly but missing cost inference function which hurts cost based balancing. Changes are:
(1) Addition of CostInferenceFunction for SplitOp
(2) Small fix in CostInferenceFunction for ConcatOp

Test Plan:
Added unit tests:

buck test //caffe2/caffe2/python/operator_test:split_op_cost_test

buck test //caffe2/caffe2/python/operator_test:concat_op_cost_test

Reviewed By: smacke

Differential Revision: D30247360

fbshipit-source-id: 989e962f3a981acc85b73aac3fb23e603b7d1591

3 years ago[docs] Merge note block in `torch.lu` documentation (#63156)
Meghan Lele [Fri, 13 Aug 2021 19:08:28 +0000 (12:08 -0700)]
[docs] Merge note block in `torch.lu` documentation (#63156)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63156

**Summary**
This commit merges the four successive `Note` blocks that appear in the
documentation for `torch.lu`. Each one only has one line in it, so all
of them have been merged into one block with a bulleted list that
contains the original items.

**Test Plan**
Continuous integration.

*Before*
<img width="888" alt="Captura de Pantalla 2021-08-12 a la(s) 10 48 39 a  m" src="https://user-images.githubusercontent.com/4392003/129244443-b7d1594e-8833-4c20-a911-e1bf7ca88a8d.png">

*After*
<img width="932" alt="Captura de Pantalla 2021-08-12 a la(s) 10 48 46 a  m" src="https://user-images.githubusercontent.com/4392003/129244462-1f39dcdb-90e0-4fd9-a95f-343b0b6be1f1.png">

**Fixes**
This commit fixes #62339.

Test Plan: Imported from OSS

Reviewed By: navahgar, pbelevich

Differential Revision: D30292633

Pulled By: SplitInfinity

fbshipit-source-id: cb9071165629bfe7316b1d2fe952e4354c75d48f

3 years ago[docs] Remove `input` parameter from `Tensor.flatten` docs (#63180)
Meghan Lele [Fri, 13 Aug 2021 18:46:54 +0000 (11:46 -0700)]
[docs] Remove `input` parameter from `Tensor.flatten` docs (#63180)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63180

**Summary**
This commit removes the `input` parameter from the signature for
`Tensor.flatten` shown in its documentation. This parameter is accepted
by `torch.flatten` but not `Tensor.flatten` (since the input is the
`Tensor` on which `flatten` is invoked).

**Test Plan**
Continuous integration.

**Fixes**
This commit fixes #57478.

Test Plan: Imported from OSS

Reviewed By: VitalyFedyunin

Differential Revision: D30293156

Pulled By: SplitInfinity

fbshipit-source-id: 4ad70d638af009fb6bdeb703433b306904d39a76

3 years ago[docs] Add cross references to `torch.transpose` and `torch.t` (#63177)
Meghan Lele [Fri, 13 Aug 2021 18:46:14 +0000 (11:46 -0700)]
[docs] Add cross references to `torch.transpose` and `torch.t` (#63177)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63177

**Summary**
This commit adds a link in the documentation for `torch.transpose` that
directs to `torch.t` and vice versa. These two functions are related and
it is useful for users of one to know about the other.

**Test Plan**
Continuous integration.

**Fixes**
This commit fixes #56267.

Test Plan: Imported from OSS

Reviewed By: VitalyFedyunin

Differential Revision: D30292654

Pulled By: SplitInfinity

fbshipit-source-id: 8e60cd7a598ff8b4756cb30141399dfe8e118338

3 years ago[docs] Mention `vsplit`, `hsplit` and `tensor_split` in Tensor views doc (#63191)
Meghan Lele [Fri, 13 Aug 2021 18:43:05 +0000 (11:43 -0700)]
[docs] Mention `vsplit`, `hsplit` and `tensor_split` in Tensor views doc (#63191)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63191

**Summary**
This commit adds `vsplit`, `hsplit` and `tensor_split` to the list of
view ops on the Tensor Views documentation page.

**Test Plan**
Continuous integration.

*Before*
<img width="195" alt="Captura de Pantalla 2021-08-12 a la(s) 2 55 07 p  m" src="https://user-images.githubusercontent.com/4392003/129275921-c1cfdf6c-9f1f-45f3-98b6-1de7a0f0cc84.png">

*After*
<img width="197" alt="Captura de Pantalla 2021-08-12 a la(s) 2 55 15 p  m" src="https://user-images.githubusercontent.com/4392003/129275936-de4afde7-0143-4e1d-b38f-c86256f4896c.png">

**Fixes**
This commit fixes #62727.

Test Plan: Imported from OSS

Reviewed By: VitalyFedyunin

Differential Revision: D30293181

Pulled By: SplitInfinity

fbshipit-source-id: 283783a4ccc3ebc50cb0a427e55c7a6cb618ffd7

3 years agoAllow Average Pooling modules to accept tensors with 0-dim batch sizes. (#62025)
Sameer Deshmukh [Fri, 13 Aug 2021 18:27:47 +0000 (11:27 -0700)]
Allow Average Pooling modules to accept tensors with 0-dim batch sizes. (#62025)

Summary:
This issue fixes a part of https://github.com/pytorch/pytorch/issues/12013, which is summarized concretely in  https://github.com/pytorch/pytorch/issues/38115.

It introduces changes and tests for allowing the Average Pooling layers to accept tensors with 0 sized batch dimensions and return meaningful results.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62025

Reviewed By: VitalyFedyunin

Differential Revision: D30303256

Pulled By: jbschlosser

fbshipit-source-id: 5f727e62a7c58d2b8bb49fcc3bd7688474917ba5

3 years ago[skip ci] fix workflow code generation (#63235)
zhouzhuojie [Fri, 13 Aug 2021 17:37:07 +0000 (10:37 -0700)]
[skip ci] fix workflow code generation (#63235)

Summary:
Fixes a clean git check with code generation introduced by https://github.com/pytorch/pytorch/pull/63148

`generated-win-vs2019-cuda10-py3.yml` was renamed as `generated-win-vs2019-cuda10.1-py3.yml`

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63235

Reviewed By: VitalyFedyunin

Differential Revision: D30306474

Pulled By: zhouzhuojie

fbshipit-source-id: cbae1ace064e360e8ca0c0e997116bdb20d54d46

3 years ago[Static Runtime] Add pass to eliminate __getitem__/DictConstruct calls (#62429)
Mike Iovine [Fri, 13 Aug 2021 17:18:03 +0000 (10:18 -0700)]
[Static Runtime] Add pass to eliminate __getitem__/DictConstruct calls (#62429)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62429

Introduce a new pass to eliminate calls to `prim::DictConstruct/aten::__getitem__`. Given a graph like this:
```
%2 : Dict = prim::DictConstruct(%key, %value)
%3 : Tensor = aten::__getitem__(%2, %key)
%4 : Tensor = op(%3)
```
This pass produces a graph like this (after dead code elimination):
```
%4 : Tensor = op(%value)
```

This optimization is applied in the static runtime.

Test Plan:
`buck test //caffe2/test:jit -- TestPeephole`

**local.forward performance summary**
About 3% runtime benefit. All `DictConstruct` calls optimized out, `__getitem__` calls reduced significantly (~50% of them are cut out)
P438354810

**local_request_only.forward performance summary**
About 14% runtime benefit. Again, all `DictConstruct` calls optimized out, 50% `__getitem__` calls removed.
P438359742

There is some variance with runtime measurements, so take these numbers with a grain of salt. Also note that the benefit does not exist in the shrunk model since there are no `DictConstruct` calls

Reviewed By: hlu1

Differential Revision: D29995087

fbshipit-source-id: f376376a46ff808115afd2d60446e5db8f6f752f

3 years agoFixing user inputs for low, high in `make_tensor` (#61108)
Kushashwa Ravi Shrimali [Fri, 13 Aug 2021 17:12:01 +0000 (10:12 -0700)]
Fixing user inputs for low, high in `make_tensor` (#61108)

Summary:
**TODOs:**

* [x] Do not clamp inputs for low and high when given and valid.
* [x] Devise rules for modifying `low` and `high` when extremals/invalid values passed.
* [x] Testing with `test_references_numerics_hard` with the revised changes. _(I've tested locally, the changes will take place in a separate PR though after offline discussion with mruberry)_
* [x] Revise comments/documentation for `make_tensor`

See https://github.com/pytorch/pytorch/issues/61758 for tracker issue.

cc: mruberry pmeier

Pull Request resolved: https://github.com/pytorch/pytorch/pull/61108

Reviewed By: VitalyFedyunin

Differential Revision: D30296167

Pulled By: mruberry

fbshipit-source-id: 67e8d15b173209a9c97ca013231494a5fa99f8c7

3 years ago[hackathon] fix benchmarking script in CONTRIBUTING (#63199)
Natalia Gimelshein [Fri, 13 Aug 2021 16:49:15 +0000 (09:49 -0700)]
[hackathon] fix benchmarking script in CONTRIBUTING (#63199)

Summary:
[skip ci]
Per title

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63199

Reviewed By: mruberry

Differential Revision: D30305487

Pulled By: ngimel

fbshipit-source-id: 2704c4f08ab976a55c9f8c2fe54cd4f3f39412cf

3 years ago[codemod][lint][caffe2] Extend BLACK coverage
Andres Suarez [Fri, 13 Aug 2021 16:26:38 +0000 (09:26 -0700)]
[codemod][lint][caffe2] Extend BLACK coverage

Test Plan: Sandcastle

Reviewed By: zsol

Differential Revision: D30302716

fbshipit-source-id: f9724d4f4d1b8950f581cc2c6c77eedf19b4b6fc

3 years agoENH Adds no_batch_dim to FractionalMaxPool2d (#62490)
Thomas J. Fan [Fri, 13 Aug 2021 15:43:04 +0000 (08:43 -0700)]
ENH Adds no_batch_dim to FractionalMaxPool2d (#62490)

Summary:
Towards https://github.com/pytorch/pytorch/issues/60585

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62490

Reviewed By: bdhirsh

Differential Revision: D30287143

Pulled By: jbschlosser

fbshipit-source-id: 1b9dd932157f571adf3aa2c98c3c6b56ece8fa6e

3 years ago[JIT] Add a flag to rethrow caught exception in jit interpreter (#63073)
Don Jang [Fri, 13 Aug 2021 15:37:23 +0000 (08:37 -0700)]
[JIT] Add a flag to rethrow caught exception in jit interpreter (#63073)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63073

It turned out that it's less than ideal to print out verbose stacktrace in exception messages in high-QPS services (see the related task) with a non-significant failure rate due to the truncation of long stacktrace which results in losing the original exception message thrown from native code. It is actually desirable to retain only the message of the original exception directly thrown from native code in such a usecase.

This change adds a new flag `torch_jit_disable_exception_stacktrace` to the pytorch jit interpreter to suppress stacktrace in the messages of exception thrown from the interpreter.

Reviewed By: Krovatkin

Differential Revision: D30241792

fbshipit-source-id: c340225c69286663cbd857bd31ba6f1736b1ac4c

3 years agoPort `norm` kernel to structured kernels. (#62711)
Yukio Siraichi [Fri, 13 Aug 2021 15:20:19 +0000 (08:20 -0700)]
Port `norm` kernel to structured kernels. (#62711)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62711

Tracking issue: #55070

Test Plan: Imported from OSS

Reviewed By: anjali411

Differential Revision: D30109866

Pulled By: ezyang

fbshipit-source-id: 894c9496894d059c7690a174b75bbd4db7ed6016

3 years agoPort `prod` kernel to structured kernels. (#62024)
Yukio Siraichi [Fri, 13 Aug 2021 15:20:19 +0000 (08:20 -0700)]
Port `prod` kernel to structured kernels. (#62024)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62024

Tracking issue: #55070

In this PR, I also broke down the meta functions of other reduction kernels (e.g. `all`,
`argmax`, `sum`) into the composition of common patterns.

Test Plan: Imported from OSS

Reviewed By: ejguan

Differential Revision: D29847122

Pulled By: ezyang

fbshipit-source-id: a6680a6cf6e59bb46b8ffe7bf2a3a611d6e0fd14

3 years agoPort `mean` kernel to structured kernels. (#61643)
Yukio Siraichi [Fri, 13 Aug 2021 15:20:19 +0000 (08:20 -0700)]
Port `mean` kernel to structured kernels. (#61643)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/61643

Tracking issue: #55070

Test Plan: Imported from OSS

Reviewed By: ejguan

Differential Revision: D29783866

Pulled By: ezyang

fbshipit-source-id: dc95baf593096c03fb5f292ee6c36de3cc7f2b35

3 years agoRemove req to call step() in training loop (#63164)
Andrew Gu [Fri, 13 Aug 2021 15:19:23 +0000 (08:19 -0700)]
Remove req to call step() in training loop (#63164)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/63164

Test Plan: Imported from OSS

Reviewed By: mrshenli

Differential Revision: D30284616

Pulled By: andwgu

fbshipit-source-id: afdb677fb08851b139178a9f6d782196f26773e1

3 years agoPass `_allow_empty_param_list` into func opt ctor (#63163)
Andrew Gu [Fri, 13 Aug 2021 15:19:23 +0000 (08:19 -0700)]
Pass `_allow_empty_param_list` into func opt ctor (#63163)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/63163

Test Plan: Imported from OSS

Reviewed By: mrshenli

Differential Revision: D30284615

Pulled By: andwgu

fbshipit-source-id: 4857f5b618ec5b007648737ab532ce605e5d70dc

3 years agoSimplify data structures, add uniform approximation, fix mem leak (#63162)
Andrew Gu [Fri, 13 Aug 2021 15:19:23 +0000 (08:19 -0700)]
Simplify data structures, add uniform approximation, fix mem leak (#63162)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/63162

Test Plan: Imported from OSS

Reviewed By: mrshenli

Differential Revision: D30284617

Pulled By: andwgu

fbshipit-source-id: 9bd9e5f89abcc0d3dac56b85d55cc88e843baa9f

3 years ago[docs][ao] update quantize_per_tensor to mention overloads (#63165)
Supriya Rao [Fri, 13 Aug 2021 14:58:38 +0000 (07:58 -0700)]
[docs][ao] update quantize_per_tensor to mention overloads (#63165)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63165

Add details about the overloads for
* list of tensors input
* supporting tensor scale/zero-point inputs

Test Plan:
CI

Imported from OSS

Reviewed By: bdhirsh

Differential Revision: D30291045

fbshipit-source-id: 9fc6418792c5e3a35417eeb8d31de4a4bfcbb7a5

3 years agoMake saved tensors default hooks thread local (#62909)
Victor Quach [Fri, 13 Aug 2021 14:47:12 +0000 (07:47 -0700)]
Make saved tensors default hooks thread local (#62909)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62909

This PR makes saved tensors default hooks thread local.
This allows using default hooks in a multithreaded context.

Test Plan: Imported from OSS

Reviewed By: albanD

Differential Revision: D30165416

Pulled By: Varal7

fbshipit-source-id: 10a7d580661d3d94bdaf398c4e076b7bea11c16b

3 years agoAllow 0-dim batch sizes for AdaptiveMaxPool and MaxPool. (#62088)
Sameer Deshmukh [Fri, 13 Aug 2021 14:31:42 +0000 (07:31 -0700)]
Allow 0-dim batch sizes for AdaptiveMaxPool and MaxPool. (#62088)

Summary:
This issue fixes a part of https://github.com/pytorch/pytorch/issues/12013, which is summarized concretely in  https://github.com/pytorch/pytorch/issues/38115.

This PR allows `MaxPool` and `AdaptiveMaxPool` to accept tensors whose batch size is 0. Some changes have been made to modernize the tests so that they will show the name of C++ function that throws an error.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62088

Reviewed By: bdhirsh

Differential Revision: D30281285

Pulled By: jbschlosser

fbshipit-source-id: 52bffc67bfe45a78e11e4706b62cce1469eba1b9

3 years agoDOC Improve documentation for LayerNorm (#63144)
AspenStars [Fri, 13 Aug 2021 13:40:41 +0000 (06:40 -0700)]
DOC Improve documentation for LayerNorm (#63144)

Summary:
In this [commit](https://github.com/pytorch/pytorch/pull/59178/commits/7026995f3ca253fbc19bf511d53f48f861799a4a) and [issue](https://github.com/pytorch/pytorch/pull/59178#issuecomment-897485295), the [Line 134](https://github.com/deniskokarev/pytorch/blob/47e286d024c183cb26a464447b34fde88b80d17d/torch/nn/modules/normalization.py#L134) will overwrite the "embedding" variate which would cause an error when initiating `nn.LayerNorm` function.

I suggest renaming the "embedding" in [Line 133](https://github.com/deniskokarev/pytorch/blob/47e286d024c183cb26a464447b34fde88b80d17d/torch/nn/modules/normalization.py#L133) to "embedding_dim".

The final example is:
```
batch, sentence_length, embedding_dim = 20, 5, 10
embedding = torch.randn(batch, sentence_length, embedding_dim)
layer_norm = nn.LayerNorm(embedding_dim)
```

Fixes #{59178}

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63144

Reviewed By: bdhirsh

Differential Revision: D30288778

Pulled By: jbschlosser

fbshipit-source-id: e74b11430e302dae5661bf6e830ee5ac6c1838c4

3 years agoRevert D30090760: [iOS] Add podspec for libTorch-lite nightly build
Alban Desmaison [Fri, 13 Aug 2021 13:36:27 +0000 (06:36 -0700)]
Revert D30090760: [iOS] Add podspec for libTorch-lite nightly build

Test Plan: revert-hammer

Differential Revision:
D30090760 (https://github.com/pytorch/pytorch/commit/e182b459d94fe77c1d9f623c94fc2621c8cc55de)

Original commit changeset: 361aa2ed24a1

fbshipit-source-id: 9c0dfee80a80eb012b142d3928204d6eb8025b0a

3 years agoOpInfo for `torch.nn.functional.normalize` (#62635)
Kushashwa Ravi Shrimali [Fri, 13 Aug 2021 13:33:40 +0000 (06:33 -0700)]
OpInfo for `torch.nn.functional.normalize` (#62635)

Summary:
See https://github.com/facebookresearch/functorch/issues/78 and https://github.com/pytorch/pytorch/issues/54261

cc: mruberry zou3519 Chillee

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62635

Reviewed By: H-Huang

Differential Revision: D30136503

Pulled By: zou3519

fbshipit-source-id: 258c069f30d9c2a51ed27dadf94f3703b9432a4a

3 years agoImplements backward for `torch.lu_solve` (#61681)
Nikita Vedeneev [Fri, 13 Aug 2021 04:15:42 +0000 (21:15 -0700)]
Implements backward for `torch.lu_solve` (#61681)

Summary:
Fixes https://github.com/pytorch/pytorch/issues/22620

Pull Request resolved: https://github.com/pytorch/pytorch/pull/61681

Reviewed By: ngimel

Differential Revision: D30063116

Pulled By: mruberry

fbshipit-source-id: e095b0cadfb7c8b37a7ef91bae5b5dc170d8ef1c

3 years agoMoving getattr_from_fqn to torch.quantization.utils (#63107)
Charles David Hernandez [Fri, 13 Aug 2021 03:57:54 +0000 (20:57 -0700)]
Moving getattr_from_fqn to torch.quantization.utils (#63107)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63107

moving this function because the functionality would be useful outside of ns
ghstack-source-id: 135727260

Test Plan: buck test //caffe2/test:quantization_fx mode/dev-nosan --keep-going --config client.id=nuclide --show-full-output -- suite

Reviewed By: supriyar

Differential Revision: D30260735

fbshipit-source-id: 58deabdd0f3b03b0ee7ee92be0548a0945084d65

3 years agoENH Migrate nll_loss2d from THC to ATen (#62826)
Thomas J. Fan [Fri, 13 Aug 2021 01:05:29 +0000 (18:05 -0700)]
ENH Migrate nll_loss2d from THC to ATen (#62826)

Summary:
Fixes https://github.com/pytorch/pytorch/issues/24608
Fixes https://github.com/pytorch/pytorch/issues/24607

With the following benchmark, the backward pass runs a little slower. This is strange since the implementation should be exactly the same.

<details>
 <summary>Benchmark script</summary>

```python
from itertools import product

import torch
import torch.nn as nn
import torch.nn.functional as F
import time

torch.manual_seed(0)
MS_PER_SECOND = 1000

def _time():
    torch.cuda.synchronize()
    return time.perf_counter() * MS_PER_SECOND

device = "cuda"
C = 3
n_runs = 30
reductions = ["none", "sum", "mean"]
Ns = [128, 256, 512]
Hs = [128, 256, 512]

for reduction, N, H in product(reductions, Ns, Hs):
    total_fwd_time = 0
    total_back_time = 0
    if reduction == "none":
        grad_out = torch.randn(N, H, H, device=device)
    else:
        grad_out = torch.randn(1)[0]

    for _ in range(n_runs):
        input = torch.randn(N, C, H, H, device=device, requires_grad=True)
        target = torch.rand(N, H, H, device=device).mul(3).floor().long()

        # forward
        start = _time()
        result = F.nll_loss(input, target, reduction=reduction)
        total_fwd_time += _time() - start

    result = F.nll_loss(input, target, reduction=reduction)
    for _ in range(n_runs):
        # backward
        start = _time()
        result.backward(grad_out, retain_graph=True)
        total_back_time += _time() - start

    fwd_avg = total_fwd_time / n_runs
    bwd_avg = total_back_time / n_runs
    print(
        f"input size({N}, {C}, {H}, {H}), reduction: {reduction}, fwd: {fwd_avg:.2f} (ms), back: {bwd_avg:.2f} (ms)"
    )

```

</details>

<details>
 <summary>master results</summary>

```
input size(128, 3, 128, 128), reduction: none, fwd: 0.34 (ms), back: 0.57 (ms)
input size(128, 3, 256, 256), reduction: none, fwd: 2.56 (ms), back: 3.85 (ms)
input size(128, 3, 512, 512), reduction: none, fwd: 14.54 (ms), back: 16.62 (ms)
input size(256, 3, 128, 128), reduction: none, fwd: 1.26 (ms), back: 1.78 (ms)
input size(256, 3, 256, 256), reduction: none, fwd: 7.07 (ms), back: 8.22 (ms)
input size(256, 3, 512, 512), reduction: none, fwd: 29.38 (ms), back: 33.29 (ms)
input size(512, 3, 128, 128), reduction: none, fwd: 3.41 (ms), back: 4.05 (ms)
input size(512, 3, 256, 256), reduction: none, fwd: 14.32 (ms), back: 16.46 (ms)
input size(512, 3, 512, 512), reduction: none, fwd: 59.20 (ms), back: 66.68 (ms)
input size(128, 3, 128, 128), reduction: sum, fwd: 0.08 (ms), back: 0.21 (ms)
input size(128, 3, 256, 256), reduction: sum, fwd: 0.21 (ms), back: 0.73 (ms)
input size(128, 3, 512, 512), reduction: sum, fwd: 0.82 (ms), back: 2.86 (ms)
input size(256, 3, 128, 128), reduction: sum, fwd: 0.12 (ms), back: 0.39 (ms)
input size(256, 3, 256, 256), reduction: sum, fwd: 0.42 (ms), back: 1.45 (ms)
input size(256, 3, 512, 512), reduction: sum, fwd: 1.53 (ms), back: 5.66 (ms)
input size(512, 3, 128, 128), reduction: sum, fwd: 0.21 (ms), back: 0.74 (ms)
input size(512, 3, 256, 256), reduction: sum, fwd: 0.78 (ms), back: 2.86 (ms)
input size(512, 3, 512, 512), reduction: sum, fwd: 2.98 (ms), back: 11.23 (ms)
input size(128, 3, 128, 128), reduction: mean, fwd: 0.07 (ms), back: 0.21 (ms)
input size(128, 3, 256, 256), reduction: mean, fwd: 0.21 (ms), back: 0.73 (ms)
input size(128, 3, 512, 512), reduction: mean, fwd: 0.82 (ms), back: 2.86 (ms)
input size(256, 3, 128, 128), reduction: mean, fwd: 0.13 (ms), back: 0.39 (ms)
input size(256, 3, 256, 256), reduction: mean, fwd: 0.42 (ms), back: 1.45 (ms)
input size(256, 3, 512, 512), reduction: mean, fwd: 1.54 (ms), back: 5.65 (ms)
input size(512, 3, 128, 128), reduction: mean, fwd: 0.22 (ms), back: 0.74 (ms)
input size(512, 3, 256, 256), reduction: mean, fwd: 0.78 (ms), back: 2.87 (ms)
input size(512, 3, 512, 512), reduction: mean, fwd: 2.98 (ms), back: 11.23 (ms)
```

</details>

<details>
 <summary>PR results</summary>

```
input size(128, 3, 128, 128), reduction: none, fwd: 0.33 (ms), back: 0.59 (ms)
input size(128, 3, 256, 256), reduction: none, fwd: 2.51 (ms), back: 3.92 (ms)
input size(128, 3, 512, 512), reduction: none, fwd: 14.52 (ms), back: 17.05 (ms)
input size(256, 3, 128, 128), reduction: none, fwd: 1.23 (ms), back: 1.85 (ms)
input size(256, 3, 256, 256), reduction: none, fwd: 7.07 (ms), back: 8.45 (ms)
input size(256, 3, 512, 512), reduction: none, fwd: 29.39 (ms), back: 34.21 (ms)
input size(512, 3, 128, 128), reduction: none, fwd: 3.40 (ms), back: 4.18 (ms)
input size(512, 3, 256, 256), reduction: none, fwd: 14.33 (ms), back: 16.90 (ms)
input size(512, 3, 512, 512), reduction: none, fwd: 59.04 (ms), back: 68.36 (ms)
input size(128, 3, 128, 128), reduction: sum, fwd: 0.07 (ms), back: 0.25 (ms)
input size(128, 3, 256, 256), reduction: sum, fwd: 0.21 (ms), back: 0.86 (ms)
input size(128, 3, 512, 512), reduction: sum, fwd: 0.82 (ms), back: 3.33 (ms)
input size(256, 3, 128, 128), reduction: sum, fwd: 0.12 (ms), back: 0.46 (ms)
input size(256, 3, 256, 256), reduction: sum, fwd: 0.42 (ms), back: 1.70 (ms)
input size(256, 3, 512, 512), reduction: sum, fwd: 1.53 (ms), back: 6.58 (ms)
input size(512, 3, 128, 128), reduction: sum, fwd: 0.21 (ms), back: 0.87 (ms)
input size(512, 3, 256, 256), reduction: sum, fwd: 0.78 (ms), back: 3.34 (ms)
input size(512, 3, 512, 512), reduction: sum, fwd: 2.98 (ms), back: 13.07 (ms)
input size(128, 3, 128, 128), reduction: mean, fwd: 0.07 (ms), back: 0.26 (ms)
input size(128, 3, 256, 256), reduction: mean, fwd: 0.21 (ms), back: 0.86 (ms)
input size(128, 3, 512, 512), reduction: mean, fwd: 0.82 (ms), back: 3.34 (ms)
input size(256, 3, 128, 128), reduction: mean, fwd: 0.12 (ms), back: 0.46 (ms)
input size(256, 3, 256, 256), reduction: mean, fwd: 0.42 (ms), back: 1.72 (ms)
input size(256, 3, 512, 512), reduction: mean, fwd: 1.53 (ms), back: 6.60 (ms)
input size(512, 3, 128, 128), reduction: mean, fwd: 0.21 (ms), back: 0.87 (ms)
input size(512, 3, 256, 256), reduction: mean, fwd: 0.78 (ms), back: 3.33 (ms)
input size(512, 3, 512, 512), reduction: mean, fwd: 2.98 (ms), back: 13.07 (ms)
```

</details>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62826

Reviewed By: bdhirsh

Differential Revision: D30282279

Pulled By: ngimel

fbshipit-source-id: 4aa0ff3f8af0632957417931d332ec486a12b52d

3 years agoadd autowrap_functions kwarg to fx.Tracer (#62106)
Alexander Soare [Fri, 13 Aug 2021 00:35:02 +0000 (17:35 -0700)]
add autowrap_functions kwarg to fx.Tracer (#62106)

Summary:
Implements feature request https://github.com/pytorch/pytorch/issues/62021

Test it out with

```python
from torch import fx
from torch import nn

def fx_int(x):
    return int(x)

class MyModule(nn.Module):
    def forward(self, x):
        return fx_int(x.shape[0] / 2)

tracer = fx.Tracer(autowrap_functions=(fx_int,))  # or remove kwarg to demonstrate symbolic trace error
tracer.trace(MyModule())
```

First time contributor, so please advise if I could have done anything to make lives easier for next time.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62106

Reviewed By: SplitInfinity, driazati

Differential Revision: D30080834

Pulled By: jamesr66a

fbshipit-source-id: 68fadf8c881ea7930e7afd62b642874010fe4903

3 years ago[fx] store Tracer class on Graph and GraphModule for package deserialization [v2...
Bradley Davis [Fri, 13 Aug 2021 00:27:08 +0000 (17:27 -0700)]
[fx] store Tracer class on Graph and GraphModule for package deserialization [v2, the re-do] (#63121)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63121

Re-introducing this diff with a small change to ignore setting Tracer classes on GraphModules when the Tracer class is defined not at module-level (prevents pickling).

Previous, reverted Pull Request: https://github.com/pytorch/pytorch/pull/62497

Reviewed By: houseroad

Differential Revision: D30252776

fbshipit-source-id: 42d2bc846e4b32d00563419c38c02b63cd0986e6

3 years agoShow warning in eager mode for empty containers (#62978)
Karol Sputo [Thu, 12 Aug 2021 22:36:29 +0000 (15:36 -0700)]
Show warning in eager mode for empty containers (#62978)

Summary:
Fixes https://github.com/pytorch/pytorch/issues/54873

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62978

Reviewed By: navahgar

Differential Revision: D30278343

Pulled By: ansley

fbshipit-source-id: ebb19f7b8a10720f2612b99a2668d1ebbc1f2d16

3 years ago[iOS] Add podspec for libTorch-lite nightly build (#62691)
Hanton Yang [Thu, 12 Aug 2021 22:32:25 +0000 (15:32 -0700)]
[iOS] Add podspec for libTorch-lite nightly build (#62691)

Summary:
The nightly pod version will be aliased with [PyTorch nightly build version](https://l.facebook.com/l.php?u=https%3A%2F%2Fgithub.com%2Fpytorch%2Fpytorch%2Fblob%2Fmaster%2F.circleci%2Fscripts%2Fbinary_populate_env.sh%23L88&h=AT3AeTpSGcz9YVeG7Lr_bweWOv8H2-kAMevglFfMslaZwgEPptNM59WdWj2ZER806rKVLNhQGM5EQcyFC_8xOq334LBo2J6YzgPW2LELkgASlA6UxP2gaD2 (https://github.com/pytorch/pytorch/commit/fa22f6303f5cf23058accf899723d6f589066ddf)Wy5mA6_lu_YlHHbEGPIU7ewJQD1 (https://github.com/pytorch/pytorch/commit/2d884f226365f94833df91de532e3a31b0db310d)aBSlOy) and [CocoaPods version specification](https://l.facebook.com/l.php?u=https%3A%2F%2Fguides.cocoapods.org%2Fusing%2Fthe-podfile.html%23specifying-pod-versions&h=AT3AeTpSGcz9YVeG7Lr_bweWOv8H2-kAMevglFfMslaZwgEPptNM59WdWj2ZER806rKVLNhQGM5EQcyFC_8xOq334LBo2J6YzgPW2LELkgASlA6UxP2gaD2 (https://github.com/pytorch/pytorch/commit/fa22f6303f5cf23058accf899723d6f589066ddf)Wy5mA6_lu_YlHHbEGPIU7ewJQD1 (https://github.com/pytorch/pytorch/commit/2d884f226365f94833df91de532e3a31b0db310d)aBSlOy), the version format of the podspect is `PyTorch version + nightly build date`, like `1.10.0.20210812`.

Usage:
1. Add `pod 'LibTorch-Lite-Nightly'` to `Podfile`
2. Run `pod install` to install the nightly built lib
3. Run `pod update` to update the lib to the latest version

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62691

Test Plan:
* Test on [TestApp](https://github.com/pytorch/pytorch/tree/master/ios/TestApp) and [HelloWorld](https://github.com/pytorch/ios-demo-app):
Podfile: `pod 'LibTorch-Lite-Nightly'`

* Test on Private Pod:
{F642106928}

Reviewed By: xta0

Differential Revision: D30090760

Pulled By: hanton

fbshipit-source-id: 361aa2ed24a11d6aced8374cb45f70f49bd5da52

3 years ago[BE] delete GHA generated workflow files before regen (#63148)
Rong Rong (AI Infra) [Thu, 12 Aug 2021 21:40:29 +0000 (14:40 -0700)]
[BE] delete GHA generated workflow files before regen (#63148)

Summary:
Unlike circle which all workflow goes in one file, GHA legacy generated files will stay silently in once's PR. e.g. when we change build_environment name and that's not ideal.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63148

Reviewed By: bdhirsh

Differential Revision: D30283382

Pulled By: walterddr

fbshipit-source-id: ffdd5bf9561dd38499052855a12ee5cf838a20b0

3 years ago[iOS][GPU] Fix the clamp shader function for x86_64 (#63062)
Tao Xu [Thu, 12 Aug 2021 20:18:42 +0000 (13:18 -0700)]
[iOS][GPU] Fix the clamp shader function for x86_64 (#63062)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63062

Pervasively, due to the need of supporting 10.0, we used a fp16 version of the clamp kernel on Metal, which didn't work well on x86_64. Since we don't need to support 10.0 anymore, we can use the fp32 version, which works both on arm64 and x86_64.
ghstack-source-id: 135536785

Test Plan:
- `buck test pp-macos`
- Op tests in the playground app

{F641013793}

Reviewed By: husthyc

Differential Revision: D30239931

fbshipit-source-id: 6ad1bf71422b537e052fbd7b7465ba8deb7ca0cf

3 years agoForbid inplace modification of a saved tensor's pack_hook input (#62717)
Victor Quach [Thu, 12 Aug 2021 19:36:38 +0000 (12:36 -0700)]
Forbid inplace modification of a saved tensor's pack_hook input (#62717)

Summary:
When using saved tensors hooks (especially default hooks),
if the user defines a `pack_hook` that modifies its input,
it can cause some surprising behavior.

The goal of this PR is to prevent future user headache by catching
inplace modifications of the input of `pack_hook` and raising an error if
applicable.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62717

Reviewed By: albanD

Differential Revision: D30255243

Pulled By: Varal7

fbshipit-source-id: 8d73f1e1b50b697a59a2849b5e21cf0aa7493b76

3 years agoUpdate CONTRIBUTING.md to remove ProcessGroupAgent (#63160)
Howard Huang [Thu, 12 Aug 2021 19:22:06 +0000 (12:22 -0700)]
Update CONTRIBUTING.md to remove ProcessGroupAgent (#63160)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/63160

Test Plan: Imported from OSS

Reviewed By: mrshenli

Differential Revision: D30284439

Pulled By: H-Huang

fbshipit-source-id: 53c31b6917ef5e2125e146fb0ed73ae3d76a8cf9

3 years agoadd use_strict_trace to tensorboard add_graph method (#63120)
Edward Wang (EcoF) [Thu, 12 Aug 2021 19:10:50 +0000 (12:10 -0700)]
add use_strict_trace to tensorboard add_graph method (#63120)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63120

FAIM returns dictionaries as the model output, which throws an error when trying to trace using add_graph. Pass in `strict` to the tracer to make this user configurable.

User post: https://fb.workplace.com/groups/pytorchLightning/permalink/1510194972650369/?comment_id=1510252919311241&reply_comment_id=1510281112641755

Test Plan: unit test

Reviewed By: Reubend

Differential Revision: D30265890

fbshipit-source-id: 58b25d9500b875a29a664aa9ef4c1e7f13631fa1

3 years agoRevert D30279364: [codemod][lint][fbcode/c*] Enable BLACK by default
Shen Li [Thu, 12 Aug 2021 18:39:31 +0000 (11:39 -0700)]
Revert D30279364: [codemod][lint][fbcode/c*] Enable BLACK by default

Test Plan: revert-hammer

Differential Revision:
D30279364 (https://github.com/pytorch/pytorch/commit/b0043072529b81276a69df29e00555333117646c)

Original commit changeset: c1ed77dfe43a

fbshipit-source-id: eab50857675c51e0088391af06ec0ecb14e2347e

3 years agoLayerNorm Support in autodiff: (#50467)
jiej [Thu, 12 Aug 2021 18:03:32 +0000 (11:03 -0700)]
LayerNorm Support in autodiff: (#50467)

Summary:
1. extend autodiff by adding entry for layer_norm in symbolic script, we now use native_layer_norm_backward
2. added backward function `layernorm_double_backward` for `native_layer_norm_backward`, preserves double backward support for LayerNorm in autodiff/ScriptModule
3. added python test to verify autodiff on layer_norm with various configuration of optional tensors; (verify the fix in https://github.com/pytorch/pytorch/issues/49430)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/50467

Reviewed By: eellison

Differential Revision: D30232864

Pulled By: jansel

fbshipit-source-id: b9c33075386aff96afff7415df9f94388bfb474a

Co-authored-by: Ryan Spring <rspring@nvidia.com>
Co-authored-by: Jie <jiej@nvidia.com>
3 years ago[codemod][lint][fbcode/c*] Enable BLACK by default
Zsolt Dollenstein [Thu, 12 Aug 2021 17:56:55 +0000 (10:56 -0700)]
[codemod][lint][fbcode/c*] Enable BLACK by default

Test Plan: manual inspection & sandcastle

Reviewed By: zertosh

Differential Revision: D30279364

fbshipit-source-id: c1ed77dfe43a3bde358f92737cd5535ae5d13c9a

3 years ago[reland] OpInfo: `adaptive_avg_pool2d` (#62935)
Kushashwa Ravi Shrimali [Thu, 12 Aug 2021 16:45:17 +0000 (09:45 -0700)]
[reland] OpInfo: `adaptive_avg_pool2d` (#62935)

Summary:
This PR is an attempt to reland https://github.com/pytorch/pytorch/pull/62704.

**What has changed?**

The op has non-deterministic behavior, hence an appropriate `gradcheck` wrapper had to be added.

cc: mruberry zou3519 heitorschueroff kshitij12345

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62935

Reviewed By: anjali411

Differential Revision: D30225095

Pulled By: zou3519

fbshipit-source-id: 644873cc21d44b19c8b68f9edff691913778de0e

3 years ago[BE] shorten CI name part2 (#63030)
Rong Rong (AI Infra) [Thu, 12 Aug 2021 15:13:01 +0000 (08:13 -0700)]
[BE] shorten CI name part2 (#63030)

Summary:
Fixes https://github.com/pytorch/pytorch/issues/62357
there's no need to specify cudnn version since they are recommended from cuda version already.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63030

Reviewed By: zhouzhuojie, driazati

Differential Revision: D30226354

Pulled By: walterddr

fbshipit-source-id: 7e2dc577810e0ce80ee27569c25a814566250ab1

3 years agoSkip zero test on windows (#63087)
Rohan Varma [Thu, 12 Aug 2021 07:37:30 +0000 (00:37 -0700)]
Skip zero test on windows (#63087)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63087

Test failed on windows unexpectedly see
https://github.com/pytorch/pytorch/issues/63086. Skip for now while we
investigate
ghstack-source-id: 135631811

Test Plan: CI

Reviewed By: ngimel

Differential Revision: D30251300

fbshipit-source-id: 8acb1ea8863c654c171fe989ac24446c321c085d

3 years agoBatchNorm: Use resize_output and empty, instead of empty_like (#63084)
Peter Bell [Thu, 12 Aug 2021 06:46:12 +0000 (23:46 -0700)]
BatchNorm: Use resize_output and empty, instead of empty_like (#63084)

Summary:
Fixes https://github.com/pytorch/pytorch/issues/62967

This lets each of the three implementations choose which memory format
to use for the output, meaning channels_last can be used in more cases.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63084

Reviewed By: saketh-are

Differential Revision: D30255740

Pulled By: ngimel

fbshipit-source-id: 48d42850952ec910b29521a1c4e530eb2b29df5e

3 years ago[quant] Make version 1 the default for get_default_qat_qconfig (#63043)
Supriya Rao [Thu, 12 Aug 2021 05:05:30 +0000 (22:05 -0700)]
[quant] Make version 1 the default for get_default_qat_qconfig (#63043)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63043

In version 1 we use the fused module/operator during QAT. Making this the default for all QAT runs going forward.

Older models saved after prepare_qat_fx can still load their state_dict into a model prepared using version 1.
The state_dict will still have the same attribute for the observer/fake_quant modules.

There may be some numerics difference between the old observer code in observer.py and the new fused module that was
re-written in C++/CUDA to perform observe + fake_quantize.

This PR also updates the test to check for the new module instead of the default FakeQuantize module.
Note: there are also some changes to make the operator work for multi-dim per-channel quantization + updated the test for that.

Test Plan:
python test/test_quantization.py TestSerialization.test_default_qat_qconfig

Imported from OSS

Reviewed By: raghuramank100

Differential Revision: D30232222

fbshipit-source-id: f3553a1926ab7c663bbeed6d574e30a7e90dfb5b

3 years agoFix sharded tensor tests. (#63054)
Pritam Damania [Thu, 12 Aug 2021 04:41:31 +0000 (21:41 -0700)]
Fix sharded tensor tests. (#63054)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63054

1) Ensure these tests are skipped in environments without any GPUs.
2) Add the test to run_test.py
ghstack-source-id: 135595698

Test Plan: waitforbuildbot

Reviewed By: wanchaol

Differential Revision: D30239159

fbshipit-source-id: 21b543ba72e8d10182bc77e7ae1fd34fd4096509

3 years agoPort `log_softmax_backward_data` to structured kernel (#62372)
Meghan Lele [Thu, 12 Aug 2021 04:01:28 +0000 (21:01 -0700)]
Port `log_softmax_backward_data` to structured kernel (#62372)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/62372

Test Plan: Imported from OSS

Reviewed By: saketh-are

Differential Revision: D30240242

Pulled By: SplitInfinity

fbshipit-source-id: 67d5e4b1543c2e43675e905ce18ca49c11e33748

3 years agoPort `log_softmax` to structured kernel (#57374)
Meghan Lele [Thu, 12 Aug 2021 04:01:28 +0000 (21:01 -0700)]
Port `log_softmax` to structured kernel (#57374)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/57374

Test Plan: Imported from OSS

Reviewed By: saketh-are

Differential Revision: D30240243

Pulled By: SplitInfinity

fbshipit-source-id: de6617c75d16e26d607a884c25b8752b7b561737

3 years agoAdd ciflow_ruleset.json generator along with gha ci (#63097)
zhouzhuojie [Thu, 12 Aug 2021 00:09:02 +0000 (17:09 -0700)]
Add ciflow_ruleset.json generator along with gha ci (#63097)

Summary:
- Add `.github/generated-ciflow-ruleset.json` for ciflow-bot (so that we can generate better comments)
- The lint job also checks git dirty to make sure that the file is always in sync with ciflow configs

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63097

Reviewed By: saketh-are

Differential Revision: D30263278

Pulled By: zhouzhuojie

fbshipit-source-id: bad68105a228e892ba071b29ecfdf433e1038054

3 years agoImprove IMethod::getArgumentNames to deal with empty argument names list (#62947)
Jiewen Tan [Wed, 11 Aug 2021 23:42:34 +0000 (16:42 -0700)]
Improve IMethod::getArgumentNames to deal with empty argument names list (#62947)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62947

This diff improved IMethod::getArgumentNames to deal with empty argument names list.

Test Plan:
buck test mode/dev //caffe2/caffe2/fb/predictor:pytorch_predictor_test -- PyTorchDeployPredictor.GetEmptyArgumentNamesValidationMode
buck test mode/dev //caffe2/caffe2/fb/predictor:pytorch_predictor_test -- PyTorchDeployPredictor.GetEmptyArgumentNamesRealMode

Reviewed By: wconstab

Differential Revision: D30179974

fbshipit-source-id: c7aec35c360a73318867c5b77ebfec3affee47e3

3 years agoFix Nnapi backend execute's dangling pointer (#63092)
Amy He [Wed, 11 Aug 2021 21:24:06 +0000 (14:24 -0700)]
Fix Nnapi backend execute's dangling pointer (#63092)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63092

Bug discovered while testing NNAPI Delegate on SparkAR.
Using
```
c10::IntArrayRef order = {0, 2, 3, 1};
fixed_inputs.push_back(tensorInp.get(i).permute(order).contiguous());
```
results in a garbage value for order in `permute()`.
Moving order inside the call to `permute()` fixes this issue. Problem is seemingly related to https://github.com/pytorch/pytorch/issues/44409, but luckily the solution in this case is simple.

Bug wasn't caught earlier, since regular unit tests weren't affected by the dangling pointer, and address sanitizer NNAPI tests are turned off due to there being a different failure (T95764916).
ghstack-source-id: 135526129

Test Plan:
Run Unit tests: `python test/test_jit.py`

Build and run SparkAR on an Android phone at the top of this diff stack (D30173959): `buck build --show-output arstudioplayer_arm64_debug -c pt.enable_nnapi=1`

Reviewed By: raziel, iseeyuan

Differential Revision: D30237504

fbshipit-source-id: c946d81feefc453b43d9295d8d6f509cafdcec03

3 years agoFix warnings (#62930)
Nikita Shulga [Wed, 11 Aug 2021 21:05:55 +0000 (14:05 -0700)]
Fix warnings (#62930)

Summary:
Add `-Wno-writable-strings`(which is clang's flavor of `-Wwrite-strings`) to list of warnings ignored while compiling torch_python.
Avoid unnecessary copies in range loop
Fix number of signed-unsigned comparisons

Found while building locally on M1

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62930

Reviewed By: albanD

Differential Revision: D30171981

Pulled By: malfet

fbshipit-source-id: 25bd43dab5675f927ca707e32737ed178b04651e

3 years ago[iOS][GPU] Consolidate array and non-array kernel for upsampling_nearest2d (#63061)
Tao Xu [Wed, 11 Aug 2021 20:28:09 +0000 (13:28 -0700)]
[iOS][GPU] Consolidate array and non-array kernel for upsampling_nearest2d (#63061)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63061

Cleanup the redundant shader code for the upsampling nearest kernel.
ghstack-source-id: 135524349

Test Plan:
- `buck test pp-macos`
- Op tests in PyTorchPlayground app

Reviewed By: husthyc

Differential Revision: D30236905

fbshipit-source-id: e1e001b446452b077e6db719b0519c9070f3300b

3 years agoirange-ify 13b (#62476)
Richard Barnes [Wed, 11 Aug 2021 20:12:16 +0000 (13:12 -0700)]
irange-ify 13b (#62476)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/62476

Test Plan: Sandcastle

Reviewed By: malfet

Differential Revision: D30001445

fbshipit-source-id: 6f4525338c80e9f929695f47f36ca9c72d96a75d

3 years agoAdd BFloat16 support for unique and unique_consecutive on CPU (#62559)
CaoE [Wed, 11 Aug 2021 19:51:28 +0000 (12:51 -0700)]
Add BFloat16 support for unique and unique_consecutive on CPU (#62559)

Summary:
Add BFloat16 support for unique and unique_consecutive on CPU.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62559

Reviewed By: saketh-are

Differential Revision: D30250675

Pulled By: ngimel

fbshipit-source-id: 26e48f971d87f3b86db237e8ad3a4b74eb3c1def

3 years agoAdd Github action to upload full source releases (#63022)
Alexander Grund [Wed, 11 Aug 2021 19:42:32 +0000 (12:42 -0700)]
Add Github action to upload full source releases (#63022)

Summary:
Those release tarballs include the submodules.
The action is run on every tag, master-branch push but will not upload anything.
This makes sure nothing is broken when an actual release happens.

On created releases the action runs and uploads the tarball

Fixes https://github.com/pytorch/pytorch/issues/62708

As I don't have access rights here and testing is obviously hard (as a new release needs to be published), I set up a test at https://github.com/Flamefire/pytorch/releases/tag/testtag
See also the run(s) at https://github.com/Flamefire/pytorch/actions/workflows/create_release.yml

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63022

Reviewed By: saketh-are

Differential Revision: D30256253

Pulled By: seemethere

fbshipit-source-id: ab5fe131452de14ae3768b91c221e68c536cb3aa

3 years agoEmbedding thrust->cub: unique (#63042)
Xiang Gao [Wed, 11 Aug 2021 19:34:58 +0000 (12:34 -0700)]
Embedding thrust->cub: unique (#63042)

Summary:
Followup of https://github.com/pytorch/pytorch/pull/62495

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63042

Reviewed By: saketh-are

Differential Revision: D30231084

Pulled By: ngimel

fbshipit-source-id: 03b0a88107e8a2aee3570881d81bf2b676f525cd

3 years ago[PyTorch] Add flop count for addmm (#61895)
Howard Cheng [Wed, 11 Aug 2021 19:32:10 +0000 (12:32 -0700)]
[PyTorch] Add flop count for addmm (#61895)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/61895

* Add FLOP count for addmm, should be `2*m*n*k`.

Share the same code path for `addmm` and `mm`.

Test Plan:
Imported from OSS

`python test/test_profiler.py`
Run a sample profile and check that FLOPS for `aten::addmm` is correct.

`[chowar@devbig053.frc2 ~/local/pytorch/build] ninja bin/test_jit`
`[chowar@devbig053.frc2 ~/local/pytorch/build] ./bin/test_jit --gtest_filter='ComputeFlopsTest*'`

Reviewed By: dskhudia

Differential Revision: D29785671

fbshipit-source-id: d1512036202d7234a981bda897af1f75808ccbfe

3 years agoXNNPack Input Pointer Caching Comment (#62818)
Salil Desai [Wed, 11 Aug 2021 18:51:58 +0000 (11:51 -0700)]
XNNPack Input Pointer Caching Comment (#62818)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62818

Added a comment to explain why we no longer need to manually cache pointers/parameters for convolution, as removed in D29777605 (https://github.com/pytorch/pytorch/commit/f5c6c3947e4618d30ebd68a414f1cfcda27bdcd4)

Test Plan: Sandcastle tests (no code changed)

Reviewed By: kimishpatel

Differential Revision: D30113489

fbshipit-source-id: d697f05816acbd367d59a4aced1925303c683d40

3 years ago`_convert_coo_to_csr` CPP and CUDA functionality (#61838)
rusty1s [Wed, 11 Aug 2021 18:35:53 +0000 (11:35 -0700)]
`_convert_coo_to_csr` CPP and CUDA functionality (#61838)

Summary:
Fixes https://github.com/pytorch/pytorch/issues/57381 and improves https://github.com/pytorch/pytorch/pull/61340 via dedicated `coo_to_csr` functionalities.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/61838

Reviewed By: ezyang

Differential Revision: D30132736

Pulled By: cpuhrsch

fbshipit-source-id: a1fd074c0d70366a524d219a620b94f8bed71d7c

3 years agoAdd a _RemoteDevice structure for ShardedTensor/ShardingSpec. (#62927)
Pritam Damania [Wed, 11 Aug 2021 18:22:48 +0000 (11:22 -0700)]
Add a _RemoteDevice structure for ShardedTensor/ShardingSpec. (#62927)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62927

As part of the ShardedTensor work, we realized we do need some sort of
_RemoteDevice structure that deals with our format of "workername/device" so
that users don't have to worry about parsing this string directly.

Right now this structure is just the bare minimum and is mostly a container for
describing a remote device. It is currently only used in ShardedTensor,
ShardingSpec and RemoteModule.

Once we actually have a consolidated remote device proposal, this class can be
extended appropriately if needed.
ghstack-source-id: 135534086

Test Plan:
1) unit tests
2) waitforbuildbot

Reviewed By: SciPioneer

Differential Revision: D30170689

fbshipit-source-id: 1ac2e81c7a597dc40bf3fbf2c1168c382c66649f

3 years ago[Pytorch Edge] Move RuntimeCompatibilityInfo Factory Method (#63005)
Jacob Szwejbka [Wed, 11 Aug 2021 18:14:25 +0000 (11:14 -0700)]
[Pytorch Edge] Move RuntimeCompatibilityInfo Factory Method (#63005)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63005

Realized I forgot to move the Runtime half of these functions be within the struct.

Test Plan: ci

Reviewed By: pavithranrao

Differential Revision: D30205521

fbshipit-source-id: ccd87d7d78450dd0dd23ba493bbb9d87be4640a5

3 years ago[easy] add an `inplace` argument to MutableNetProto.to_net() and core.Net() construct...
Stephen Macke [Wed, 11 Aug 2021 18:09:02 +0000 (11:09 -0700)]
[easy] add an `inplace` argument to MutableNetProto.to_net() and core.Net() constructor (#63068)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63068

The caffe2 core.Net constructor can accept a caffe2_pb2.NetDef proto, but it always creates a copy. This is wasteful when we can prove that the proto being passed to it will not be used anywhere else. So we add an "inplace" argument to the `core.Net` constructor that allows clients to give away ownership of the passed proto without copying. We default this argument to `False`, ensuring that behavior does not change unless explicitly requested.

Test Plan: Let CI run.

Differential Revision: D29976510

fbshipit-source-id: 26e13ca76f3431b8ef0de51f08bbf263491d323e

3 years agoFix gha render-test-result mixed failure passthrough (#63056)
zhouzhuojie [Wed, 11 Aug 2021 16:42:15 +0000 (09:42 -0700)]
Fix gha render-test-result mixed failure passthrough (#63056)

Summary:
To fix something like https://github.com/pytorch/pytorch/actions/runs/1114555082

![image](https://user-images.githubusercontent.com/658840/128956528-86997457-5e18-4ae1-83cc-aa7d0ca03c0e.png)

Not sure why `needs.test.result` doesn't capture the `failure` case before, so changed it to `needs.test.result != 'skipped' || failure()`

Pull Request resolved: https://github.com/pytorch/pytorch/pull/63056

Reviewed By: walterddr, tktrungna

Differential Revision: D30240112

Pulled By: zhouzhuojie

fbshipit-source-id: d159cc3f79ed5d604ae12583736b37ac28e8d87c

3 years agoFix issues with printing certain torch modules (#62447)
Yida Wang [Wed, 11 Aug 2021 16:36:49 +0000 (09:36 -0700)]
Fix issues with printing certain torch modules (#62447)

Summary:
Fixes https://github.com/pytorch/pytorch/issues/54420

When I tested on master, with the testing code, there were multiple objects on the garbage collector that cannot be printed.

Testing code:
```
import torch
import gc
import os
import sys

print(torch.__version__)

a = torch.rand(10)

print(a)

objects = gc.get_objects()

for i in range(len(objects)):
   print(objects[i])
```

### 1
```
print(torch.classes)
```

Like SplitInfinity has mentioned in the GitHub issue, the solution here is to set `__file__` for `torch.classes` to something. Similar to [_ops.py](https://github.com/pytorch/pytorch/blob/master/torch/_ops.py#L69), where `__file__` is set to `_ops.py`, we could set `__file__` for torch.classes to `_classes.py`.

### 2
```
print(torch._ops.ops.quantized)
print(torch._ops.ops.atan)
```

When we try to print these two modules, it will call `_OpNamespace::__getattr__`, but the `op_name` is `__file__`. This becomes a problem when `torch._C._jit_get_operation(qualified_op_name)` [(link)](https://github.com/pytorch/pytorch/blob/master/torch/_ops.py#L60) tries to look for an actual op on the native C++ side.

Only when we get the attribute for an actual op, e.g. `print(torch._ops.ops.quantized.elu)`, the `op_name` becomes proper (e.g. `elu`).

My current solution is to return a hardcoded string (i.e. “torch.ops”) if `op_name` is `"__file__"`.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62447

Reviewed By: saketh-are

Differential Revision: D30234654

Pulled By: yidawang-oss

fbshipit-source-id: de43a8f599739c749fb3307eea015cc61f1da60e

3 years agoShard python_functions.cpp (#62186)
Peter Bell [Wed, 11 Aug 2021 15:44:08 +0000 (08:44 -0700)]
Shard python_functions.cpp (#62186)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62186

This file takes 6 minutes on its own to compile and is the limiting factor for
building `libtorch_python` on a 32-core threadripper. This splits the file into
5 shards which take around 50 seconds each to compile.

Test Plan: Imported from OSS

Reviewed By: bdhirsh

Differential Revision: D29962046

Pulled By: albanD

fbshipit-source-id: df13cfaebd54296f10609f67ae74a850c329bd37

3 years agoFix inconsisteny between Python and JIT power operation (#62842)
Sze Wai Celeste Yuen [Wed, 11 Aug 2021 15:38:13 +0000 (08:38 -0700)]
Fix inconsisteny between Python and JIT power operation (#62842)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/62842

Test Plan:
Wrote unit test TestAtenPow to test behavior of aten::pow when:
1. base is int, exponent is int
2. base is int, exponent is float
3. base is float, exponent is int
4. base is float, exponent is float

Specifically, we test that when base is zero and exponent is negative, we raise error. In all other cases, we expect behavior to be the same as the result returned by Python.

It is because the cpp code relies on overloading, we need to make sure all combinations of types give us the expected result.

Reviewed By: zhxchen17

Differential Revision: D30146115

Pulled By: szewaiyuen7

fbshipit-source-id: dc661897ad38da286ee454120fbe41314b7f2995

3 years agoFix CUDA_KERNEL_ASSERT ambiguous symbol in NDEBUG mode (#62527)
Dmytro Dzhulgakov [Wed, 11 Aug 2021 08:08:45 +0000 (01:08 -0700)]
Fix CUDA_KERNEL_ASSERT ambiguous symbol in NDEBUG mode (#62527)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62527

If NDEBUG is applied inconsistently in compilation we might get 'ambiguous declaration' error. Let's make sure that the forward declaration matches glibc including all specifiers.

Test Plan: sandcastle

Reviewed By: mdschatz

Differential Revision: D30030051

fbshipit-source-id: 9f4d5f1d4e74f0a4eaeeaaaad76b93ee485d8bcd

3 years ago[4/N] Enable opt-asan for distributed unit tests. (#62051)
Pritam Damania [Wed, 11 Aug 2021 05:37:14 +0000 (22:37 -0700)]
[4/N] Enable opt-asan for distributed unit tests. (#62051)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62051

The goal here is to enable opt-asan for "spawn" based unit tests since
this works for "spawn" unlike "dev-asan". As a result, we can run ASAN for
"spawn" unit tests as well.

This means we can completely remove fork unit tests from the code base since
the only purpose for these tests was to run ASAN.
ghstack-source-id: 135523770

Test Plan: waitforbuildbot

Reviewed By: SciPioneer

Differential Revision: D29854514

fbshipit-source-id: 02a5bfcfae2afc21badecff77082c7a6ad83636b

3 years agoBack out "[fx] store Tracer class on Graph and GraphModule for package deserializatio...
Lu Fang [Wed, 11 Aug 2021 04:56:41 +0000 (21:56 -0700)]
Back out "[fx] store Tracer class on Graph and GraphModule for package deserialization" (#63053)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63053

Original commit changeset: eca09424ad30

The original diff - D30019214 (https://github.com/pytorch/pytorch/commit/6286d338785c48a3e7a9b969e2bc3bd4d502851d) breaks the publish flow in model saving.

Test Plan: ci

Differential Revision: D30236517

fbshipit-source-id: 3e05db02fc1cbbc2ed262c83bf56d555277abb34

3 years agorebase for autocast updates to include device_type and dtype flags (#61002)
Rishi Puri [Wed, 11 Aug 2021 03:02:07 +0000 (20:02 -0700)]
rebase for autocast updates to include device_type and dtype flags (#61002)

Summary:
Fixes #{55374}
https://github.com/pytorch/pytorch/issues/55374

Pull Request resolved: https://github.com/pytorch/pytorch/pull/61002

Reviewed By: malfet, mruberry

Differential Revision: D30016812

Pulled By: ngimel

fbshipit-source-id: 6e09a29f539d28e9aea5cd9489b1e633cc588033

3 years agoFix missing element types and shapes when autograd.Function has multiple tensor outpu...
Wei-Sheng Chin [Wed, 11 Aug 2021 02:46:46 +0000 (19:46 -0700)]
Fix missing element types and shapes when autograd.Function has multiple tensor outputs (#57966)

Summary:
When generating IR for autograd.Function, if the function has multiple outputs, a TupleUnpack may be inserted after the original function node, and Pytorch only assigns proper information (tensor element type and shape) to the TupleUnpack and forgets the original function node. In contrast, if autograd.Function only produces one output, the original function node may have tensor
element type and shape in its output schema.

Before this PR:
- (simplified) IR for autograd.Function with one output: input (tensor, dtype=float32, shape=[2, 3]) -> PythonOp -> output (tensor, dtype=float32, shape=[4, 5])
- (simplified) IR for autograd.Function with one output: input (tensor, dtype=float32, shape=[2, 3]) -> PythonOp -> output_0 **(tensor)**, output_1 **(tensor)** -> TupleUnpack output_2 (tensor, dtype=float32, shape=[4, 5]), output_3 (tensor, dtype=float32, shape=[6, 7])

After this PR:
- (simplified) IR for autograd.Function with one output: input (tensor, dtype=float32, shape=[2, 3]) -> PythonOp -> output (tensor, dtype=float32, shape=[4, 5])
- (simplified) IR for autograd.Function with one output: input (tensor, dtype=float32, shape=[2, 3]) -> PythonOp ->output_0 **(tensor, dtype=float32, shape=[4, 5])**, output_1 **(tensor, dtype=float32, shape=[6, 7])** -> TupleUnpack output_2 (tensor, dtype=float32, shape=[4, 5]), output_3 (tensor, dtype=float32, shape=[6, 7])

Pull Request resolved: https://github.com/pytorch/pytorch/pull/57966

Reviewed By: zhxchen17

Differential Revision: D30208207

Pulled By: gmagogsfm

fbshipit-source-id: 42a3d1f9c0932133112a85df0c49cf4ea0afa175

3 years agoremove dead code (#63031)
Natalia Gimelshein [Wed, 11 Aug 2021 01:39:45 +0000 (18:39 -0700)]
remove dead code (#63031)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/63031

Reviewed By: mruberry

Differential Revision: D30225094

Pulled By: ngimel

fbshipit-source-id: 3666a0fa120bea85225cd3ee04f89d64952d2862

3 years agoRevert D30199482: [pytorch][PR] Add BFloat16 support for unique and unique_consecutiv...
Natalia Gimelshein [Wed, 11 Aug 2021 01:23:00 +0000 (18:23 -0700)]
Revert D30199482: [pytorch][PR] Add BFloat16 support for unique and unique_consecutive on CPU

Test Plan: revert-hammer

Differential Revision:
D30199482 (https://github.com/pytorch/pytorch/commit/fc0b8e60337ae46b90ed5d2f6d1f623f0f8d6581)

Original commit changeset: 6f2d9cc1a528

fbshipit-source-id: 39e9f202bcbd978525f792173d4f97b5b329b5b1

3 years agoUse `const auto` with irange (#62990)
Richard Barnes [Wed, 11 Aug 2021 00:57:22 +0000 (17:57 -0700)]
Use `const auto` with irange (#62990)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/62990

Test Plan: Sandcastle

Reviewed By: zhouzhuojie

Differential Revision: D30199748

fbshipit-source-id: 284b208ffa3c6c4749e5ac9b1fccb28914590f2c

3 years agochange nccl version reporting (#62916)
Eddie Yan [Wed, 11 Aug 2021 00:44:40 +0000 (17:44 -0700)]
change nccl version reporting (#62916)

Summary:
https://github.com/pytorch/pytorch/issues/62295

Previously the packing and unpacking of the NCCL version "integer" was done to have parity with the upstream NCCL version encoding. However, there doesn't seem to be any place where this integer is directly compared with a version integer sourced from upstream NCCL, and syncing the encoding seems to be error-prone (e.g., a recent change where a special case was added for minor versions >= 10 https://github.com/NVIDIA/nccl/blob/7e515921295adaab72adf56ea71a0fafb0ecb5f3/src/nccl.h.in#L22).

This patch changes the reporting to return a tuple of version numbers instead (to preserve ease-of-use for comparisons) and tweaks the passing between C/Python to avoid the digit overflow problem.

CC ngimel mcarilli

Pull Request resolved: https://github.com/pytorch/pytorch/pull/62916

Reviewed By: anjali411

Differential Revision: D30201069

Pulled By: mrshenli

fbshipit-source-id: 2e4e7c69f001c3f22bd04aa6df6a992e538bea45

3 years agoUpdate test_torch_deploy (#62838)
tktrungna [Tue, 10 Aug 2021 23:24:57 +0000 (16:24 -0700)]
Update test_torch_deploy (#62838)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62838

Fixes #62380

* update test functions to call wheel install folder {sitepackages}/torch instead of build/ folder
* add symbolic link for shared libraries which are called by the tests (this is a bit hacky and should be fixed the rpath before compiling -- similar to https://github.com/pytorch/pytorch/blob/master/.jenkins/pytorch/test.sh#L204-L208).

### Test plan
check if all ci workflows pass

Test Plan: Imported from OSS

Reviewed By: walterddr

Differential Revision: D30193141

Pulled By: tktrungna

fbshipit-source-id: 72c2bd3a740fca0f72e4803df505240193692c44

3 years agoupdate test_libtorch (#62797)
tktrungna [Tue, 10 Aug 2021 23:24:57 +0000 (16:24 -0700)]
update test_libtorch (#62797)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62797

Fixes #62380

* update test functions to call wheel install folder {sitepackages}/torch instead of build/ folder
* add symbolic link for shared libraries which are called by the tests (this is a bit hacky and should be fixed the rpath before compiling -- similar to https://github.com/pytorch/pytorch/blob/master/.jenkins/pytorch/test.sh#L204-L208).

### Test plan
check if all ci workflows pass

Test Plan: Imported from OSS

Reviewed By: walterddr

Differential Revision: D30193140

Pulled By: tktrungna

fbshipit-source-id: d8e54c403f42abbbbe4556abf40c22a7955df737

3 years agoupdate test distributed (#62796)
tktrungna [Tue, 10 Aug 2021 23:24:57 +0000 (16:24 -0700)]
update test distributed (#62796)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62796

Fixes #62380

* update test functions to call wheel install folder {sitepackages}/torch instead of build/ folder
* add symbolic link for shared libraries which are called by the tests (this is a bit hacky and should be fixed the rpath before compiling -- similar to https://github.com/pytorch/pytorch/blob/master/.jenkins/pytorch/test.sh#L204-L208).

### Test plan
check if all ci workflows pass

Test Plan: Imported from OSS

Reviewed By: driazati

Differential Revision: D30193142

Pulled By: tktrungna

fbshipit-source-id: 1247f9eda1c11c763c31c7383c77545b1ead1a60

3 years agoupdate test_vulkan (#62795)
tktrungna [Tue, 10 Aug 2021 23:24:57 +0000 (16:24 -0700)]
update test_vulkan (#62795)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/62795

Test Plan: Imported from OSS

Reviewed By: driazati

Differential Revision: D30124421

Pulled By: tktrungna

fbshipit-source-id: 235ba166b02f7334e89cb2493024067851bf5b9b

3 years agoupdate test_rpc (#62781)
tktrungna [Tue, 10 Aug 2021 23:24:57 +0000 (16:24 -0700)]
update test_rpc (#62781)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/62781

Test Plan: Imported from OSS

Reviewed By: walterddr, zhouzhuojie

Differential Revision: D30124391

Pulled By: tktrungna

fbshipit-source-id: 99c275d6c9f23b4f274fd0ca19a16879ed27afd5

3 years ago[ONNX] add support for prim::Unitialized in lower_tuples pass (#56912)
Matej Sladek [Tue, 10 Aug 2021 23:19:39 +0000 (16:19 -0700)]
[ONNX] add support for prim::Unitialized in lower_tuples pass (#56912)

Summary:
Fixes https://github.com/pytorch/pytorch/issues/56911

Code from issue generates this Torchscript:
```
graph(%self : __torch__.MyModule,
      %t.1 : Tensor):
  %12 : None = prim::Constant()
  %7 : str = prim::Constant[value="Negative input"]() # /mnt/nvdl/usr/msladek/notes/python_code/unitialized.py:11:28
  %3 : int = prim::Constant[value=0]() # /mnt/nvdl/usr/msladek/notes/python_code/unitialized.py:10:15
  %9 : int = prim::Constant[value=5]() # /mnt/nvdl/usr/msladek/notes/python_code/unitialized.py:13:31
  %33 : (Tensor, Tensor) = prim::Uninitialized()
  %4 : Tensor = aten::lt(%t.1, %3) # /mnt/nvdl/usr/msladek/notes/python_code/unitialized.py:10:11
  %6 : bool = aten::Bool(%4) # /mnt/nvdl/usr/msladek/notes/python_code/unitialized.py:10:11
  %34 : (Tensor, Tensor) = prim::If(%6) # /mnt/nvdl/usr/msladek/notes/python_code/unitialized.py:10:8
    block0():
       = prim::RaiseException(%7) # /mnt/nvdl/usr/msladek/notes/python_code/unitialized.py:11:12
      -> (%33)
    block1():
      %11 : int[] = prim::ListConstruct(%9)
      %16 : Tensor = aten::zeros(%11, %12, %12, %12, %12) # /mnt/nvdl/usr/msladek/notes/python_code/unitialized.py:13:19
      %18 : int[] = prim::ListConstruct(%9)
      %23 : Tensor = aten::zeros(%18, %12, %12, %12, %12) # /mnt/nvdl/usr/msladek/notes/python_code/unitialized.py:13:35
      %24 : (Tensor, Tensor) = prim::TupleConstruct(%16, %23)
      -> (%24)
  return (%34)
```

Problem is that onnx exporter during lower_tuples pass doesn't support forwarding of tuples in prim::Unitialized.
Solution is:
1. add prim::Unitialized to supported_op in lower_tuples pass
1. As prim::Unitialized has now multiple outputs, we should call giveFreshAlias for every output

Pull Request resolved: https://github.com/pytorch/pytorch/pull/56912

Reviewed By: nikithamalgifb

Differential Revision: D29837200

Pulled By: SplitInfinity

fbshipit-source-id: 321fae6fe52b1523df5653dbb9ea73b998ef1cda

3 years agoRemove process_group_agent and faulty_process_group_agent files (#62985)
Howard Huang [Tue, 10 Aug 2021 22:56:18 +0000 (15:56 -0700)]
Remove process_group_agent and faulty_process_group_agent files (#62985)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62985

Remove the process_group_agent and faulty_process_group_agent code now that PROCESS_GROUP backend has been deprecated for RPC (https://github.com/pytorch/pytorch/issues/55615). Discussed with xush6528 that it was okay to remove ProcessGroupAgentTest and ProcessGroupAgentBench which depended on process_group_agent.

Test Plan: CI tests

Reviewed By: pritamdamania87

Differential Revision: D30195576

fbshipit-source-id: 8b4381cffadb868b19d481198015d0a67b205811