platform/upstream/pytorch.git
5 years agoQTensor (#18230)
Jerry Zhang [Wed, 3 Apr 2019 20:13:26 +0000 (13:13 -0700)]
QTensor (#18230)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18230

Implementing minimum qtensor API to unblock other workstreams in quantization

Changes:
- Added Quantizer which represents different quantization schemes
- Added qint8 as a data type for QTensor
- Added a new ScalarType QInt8
- Added QTensorImpl for QTensor
- Added following user facing APIs
  - quantize_linear(scale, zero_point)
  - dequantize()
  - q_scale()
  - q_zero_point()

Reviewed By: dzhulgakov

Differential Revision: D14524641

fbshipit-source-id: c1c0ae0978fb500d47cdb23fb15b747773429e6c

5 years agoEnforce import order to make protobuf cpp implementation in python work (#18560)
Dmytro Dzhulgakov [Wed, 3 Apr 2019 20:12:28 +0000 (13:12 -0700)]
Enforce import order to make protobuf cpp implementation in python work (#18560)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18560

We have to import python protobuf here **before** we load cpp extension.
Otherwise it breaks under certain build conditions if cpp implementation of
protobuf is used. Presumably there's some registry in protobuf library and
python side has to initialize the dictionary first, before static
initialization in python extension does so. Otherwise, duplicated protobuf
descriptors will be created and it can lead to obscure errors like

  Parameter to MergeFrom() must be instance of same class: expected caffe2.NetDef got caffe2.NetDef.

I think it also fixes https://github.com/facebookarchive/caffe2/issues/1573

Reviewed By: ezyang, iroot900

Differential Revision: D14622054

fbshipit-source-id: 2499eb88ecdee85ff8d845859048f7ae5da2a480

5 years agoPin onnx ir_version to 4 (#18768)
Lu Fang [Wed, 3 Apr 2019 20:08:21 +0000 (13:08 -0700)]
Pin onnx ir_version to 4 (#18768)

Summary:
to make test_operators.py more stable. in future, we will bump this up manually, and I think it's acceptable, since ir_version should be bumped too often.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18768

Reviewed By: zrphercule

Differential Revision: D14741514

Pulled By: houseroad

fbshipit-source-id: 0369dbc55424e345a113e49fc104a441ea290d58

5 years agofix nccl compilation to make sure it compiles for architectures that pytorch compiles...
Soumith Chintala [Wed, 3 Apr 2019 19:44:35 +0000 (12:44 -0700)]
fix nccl compilation to make sure it compiles for architectures that pytorch compiles for (#18739)

Summary:
resubmit of https://github.com/pytorch/pytorch/pull/18704 with additional fixes

Fixes https://github.com/pytorch/pytorch/issues/18359
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18739

Differential Revision: D14737274

Pulled By: soumith

fbshipit-source-id: cfbbbf68b098594bd045861d1b2c085da693ea51

5 years agopush magma init into lazyInitCUDA (#18527)
Soumith Chintala [Wed, 3 Apr 2019 19:27:19 +0000 (12:27 -0700)]
push magma init into lazyInitCUDA (#18527)

Summary:
Tries to fix C++ API's usage of MAGMA-based functions.

Attempts to Fix https://github.com/pytorch/pytorch/issues/18074
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18527

Differential Revision: D14691694

Pulled By: soumith

fbshipit-source-id: dd04e74418e486d73ea4a92193ddf79352ed71ba

5 years agoFor some files that are touched by the QTensor diff (#18765)
Jerry Zhang [Wed, 3 Apr 2019 19:01:57 +0000 (12:01 -0700)]
For some files that are touched by the QTensor diff (#18765)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18765

att

Reviewed By: ZolotukhinM

Differential Revision: D14733442

fbshipit-source-id: 525002034e6dccc2045da645e1193671fd0474b3

5 years agoFix contiguous AD and Autogradzero inconsistency (#18633)
Wanchao Liang [Wed, 3 Apr 2019 18:18:05 +0000 (11:18 -0700)]
Fix contiguous AD and Autogradzero inconsistency (#18633)

Summary:
Fixes #17962
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18633

Differential Revision: D14700449

Pulled By: wanchaol

fbshipit-source-id: 3d15d67c01b69b28394a0f2f001db90ed9fd31dc

5 years agoAdded indexing for bool tensors and bool Indices (#18583)
Iurii Zdebskyi [Wed, 3 Apr 2019 17:53:11 +0000 (10:53 -0700)]
Added indexing for bool tensors and bool Indices (#18583)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18583
ghimport-source-id: 2b1941449827f4ab632fa0f5c8cf0791a6be0845

Stack from [ghstack](https://github.com/ezyang/ghstack):
* **#18583 Added indexing for bool tensors and bool Indices**
* #18505 Added numpy conversion
* #18166 Bool Tensor for CUDA

-----------
This PR enables bool tensor indexing and indexing with bool indices. This is a part of Bool Tensor feature implementation work. The whole plan looks like this:
1. Storage Implementation [Done]
2. Tensor Creation.
    a) CPU [Done]
    b) CUDA [In review]
3. Tensor Conversions. [In review]
4. Tensor Indexing. [This PR]
5. Tensor Operations.
6. Back compatibility related changes.

TODO:
as a follow up, we should move nonzero method from TH to Aten to make code cleaner.

Change:
```
v = torch.tensor([True, False, True], dtype=torch.bool)
boolIndices = torch.tensor([True, False, False], dtype=torch.bool)
v[boolIndices]
-> tensor([True], dtype=torch.bool)

v = torch.randn(5, 7, 3)
boolIndices = torch.tensor([True, False, True, True, False], dtype=torch.bool)
v[boolIndices]
->
tensor([[[ 0.5885, -0.3322,  0.7388],
         [ 1.1182,  0.7808, -1.1492],
         [-0.7952,  0.5255, -0.0251],
         [ 0.7128,  0.8099,  1.2689],
         [-0.7018, -1.4733, -0.3732],
         [ 0.4503,  0.4986, -1.1605],
         [ 0.3348, -1.3767, -0.2976]],

        [[-2.0303, -0.4720, -0.1448],
         [-0.1914, -0.6821,  2.0061],
         [-1.0420, -0.1872, -0.3438],
         [ 1.7587, -0.4183, -0.7577],
         [ 1.0094, -0.1950, -0.2430],
         [ 0.1174,  0.3308, -0.5700],
         [ 0.1110, -0.2714,  1.3006]],

        [[-0.1946, -1.4747, -0.4650],
         [-1.0567,  1.0110, -0.2809],
         [ 0.3729, -0.5699,  0.0815],
         [-0.7733, -0.8316,  0.1674],
         [ 1.2000, -0.3745, -1.1679],
         [ 1.7105,  0.9851, -0.1907],
         [-1.1077,  0.2086, -0.0548]]])
```

Differential Revision: D14673403

fbshipit-source-id: 2b88ec2c7eb26a4f5ef64f8707fb68068d476fc9

5 years agoadd an assertion to check the param num (#18145)
Lu Fang [Wed, 3 Apr 2019 17:51:41 +0000 (10:51 -0700)]
add an assertion to check the param num (#18145)

Summary:
Introduce this check to see whether it will break any existing workflow
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18145

Reviewed By: dzhulgakov

Differential Revision: D14511711

Pulled By: houseroad

fbshipit-source-id: a7bb6ac84c9133fe94d3fe2f1a8566faed14a136

5 years agoadd Android NDK param to CI docker build script (#18782)
Jiakai Liu [Wed, 3 Apr 2019 17:42:30 +0000 (10:42 -0700)]
add Android NDK param to CI docker build script (#18782)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18782
ghimport-source-id: 6c4bde7dc835b59209c1d5f7b243f00c9fe99de2

Stack from [ghstack](https://github.com/ezyang/ghstack):
* **#18782 [pytorch] add Android NDK param to CI docker build script**

Inspired by discussion: https://github.com/pytorch/pytorch/pull/16242

Reviewed By: dreiss

Differential Revision: D14739471

fbshipit-source-id: 0a081045186cbf359eb3cdadee722741cd8cd62f

5 years agoUpgrade mkldnn-bridge for dnnlowp support (#16308)
Gu, Jinghui [Wed, 3 Apr 2019 17:29:19 +0000 (10:29 -0700)]
Upgrade mkldnn-bridge for dnnlowp support (#16308)

Summary:
The mkldnn-bridge is upgraded in this PR to support DNNLOWP operators.
Meanwhile, APIs have been updated in caffe2 to use latest version.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/16308

Differential Revision: D14697018

Pulled By: yinghai

fbshipit-source-id: ca952589098accb08295fd5aa92924c61e74d69c

5 years agoadd 'abs' builtin
Michael Kösel [Wed, 3 Apr 2019 17:11:33 +0000 (10:11 -0700)]
add 'abs' builtin

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/18502

Differential Revision: D14750173

Pulled By: eellison

fbshipit-source-id: 359cf08938ada442ca1a3b3ea14022ce10229499

5 years agoFix dense Embedding to work with double backward (#9078)
kshitij12345 [Wed, 3 Apr 2019 16:16:29 +0000 (09:16 -0700)]
Fix dense Embedding to work with double backward (#9078)

Summary:
Fixes : #6469

1. `ATen/native/native_functions.yml` had [dispatch](https://github.com/pytorch/pytorch/blob/03e7953a98875c0164cb8e2c19b45800e85f4347/aten/src/ATen/native/native_functions.yaml#L451-L455) variants for for `embedding_dense_backward` , however `embedding_backward` explicitly made [call](https://github.com/pytorch/pytorch/blob/03e7953a98875c0164cb8e2c19b45800e85f4347/aten/src/ATen/native/Embedding.cpp#L35-L45) to it, thus leading to error.

2. In case of CUDA type tensor, the function crashed used to crash on dereferencing of indices's data [pointer](https://github.com/pytorch/pytorch/blob/03e7953a98875c0164cb8e2c19b45800e85f4347/aten/src/ATen/native/Embedding.cpp#L93).

Both have been solved and checked against (on CUDA and CPU)

1.  As mentioned in the issue
```
import torch

class Test(torch.nn.Module):

    def __init__(self):
        super(Test,self).__init__()
        self.embd = torch.nn.Embedding(1000, 100)
        self.dense = torch.nn.Linear(100, 1)

    def forward(self, inp):
        inp = self.embd(inp)
        return self.dense(inp)

test = Test()
inp = torch.tensor([0,1,2,1,1])
out = test(inp)
raw_loss = out.mean(dim=0)

loss_grad = torch.autograd.grad(outputs=raw_loss,
                         inputs=list(test.parameters()),
                         retain_graph=True, create_graph=True, only_inputs=True)
norm = sum([param.norm()**2 for param in loss_grad])
loss = raw_loss + norm

loss.backward(retain_graph=True)

print(test.embd.weight.grad)

```

2. Test Script
```
import torch
import time
start = time.time()
l = [1,1]*100
input = torch.tensor([[1,0],[1,0]],device='cpu')
embedding_matrix = torch.tensor([[1.0,3.0],[2.0,4]],requires_grad=True,device='cpu')

sq = embedding_matrix * embedding_matrix
emb = torch.nn.functional.embedding(input, sq,scale_grad_by_freq=False)

print('Embedding Matrix')
print(embedding_matrix)
print('-----------------')

sum_ = emb.sum()#prod.sum()

loss_grad, = torch.autograd.grad(outputs=sum_,inputs=embedding_matrix,create_graph=True)

print('Gradient')
print(loss_grad)
print('-----------------')

sum2_ = sum_ + loss_grad.sum()
print(sum2_)
sum2_.backward()

print(embedding_matrix.grad)
print(time.time() - start)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/9078

Reviewed By: ezyang

Differential Revision: D14691901

Pulled By: soumith

fbshipit-source-id: 78e2612ba39080be564c876311671eb5a0119a0f

5 years agoHighlight NCCL all_reduce and all_gather requirements (#18741)
Shen Li [Wed, 3 Apr 2019 16:06:09 +0000 (09:06 -0700)]
Highlight NCCL all_reduce and all_gather requirements (#18741)

Summary:
See #18689
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18741

Differential Revision: D14726874

Pulled By: mrshenli

fbshipit-source-id: a92404c653e3c62fc23fa3ccacfb3b2959b2e307

5 years agoUpdating submodules
svcscm [Wed, 3 Apr 2019 15:25:14 +0000 (08:25 -0700)]
Updating submodules

Reviewed By: zpao

fbshipit-source-id: ea0b06ce68d3fd6092eaea7c835a8b51c1120ea0

5 years agoMake it possible for users for select /Zi or /ZI over /Z7 when using MSVC (#18790)
peter [Wed, 3 Apr 2019 15:19:45 +0000 (08:19 -0700)]
Make it possible for users for select /Zi or /ZI over /Z7 when using MSVC (#18790)

Summary:
Fixes https://github.com/pytorch/pytorch/issues/18701.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18790

Differential Revision: D14748195

Pulled By: ezyang

fbshipit-source-id: e50df1b5ca199a88d7b5ea3ea45d25d23cd31a27

5 years agouse optimization in D14020675 (#16945)
Jongsoo Park [Wed, 3 Apr 2019 14:55:02 +0000 (07:55 -0700)]
use optimization in D14020675 (#16945)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/16945

As title

Reviewed By: jianyuh

Differential Revision: D14020769

fbshipit-source-id: fc0f05fcc57bfe9b4aa0c5750060d7b2ba57dd7a

5 years agoAdd device and dtype to storage. (#18749)
Gregory Chanan [Wed, 3 Apr 2019 14:52:54 +0000 (07:52 -0700)]
Add device and dtype to storage. (#18749)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18749
ghimport-source-id: 9026a037f5e11cdb9ccd386f4b6b5768b9c3259b

Stack from [ghstack](https://github.com/ezyang/ghstack):
* #18751 Disallow changing the device of a tensor via set_.
* #18750 Use non-legacy constructors for tensor deserialization.
* **#18749 Add device and dtype to storage.**

The goal here is to fix our serialization, which currently depends on the legacy constructors.  Having dtype and device on Storage allows us to use the non-legacy constructors.

This fits somewhat along our goal of removing Storage, my having Storage act like a Tensor.

Differential Revision: D14729516

fbshipit-source-id: bf4a3e8669ad4859931f4a3fa56df605cbc08dcb

5 years agoUse non-legacy constructors for tensor deserialization. (#18750)
Gregory Chanan [Wed, 3 Apr 2019 14:51:15 +0000 (07:51 -0700)]
Use non-legacy constructors for tensor deserialization. (#18750)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18750
ghimport-source-id: f1475cfb67841c41d9867d4429ba9125d5c7dd07

Stack from [ghstack](https://github.com/ezyang/ghstack):
* #18751 Disallow changing the device of a tensor via set_.
* **#18750 Use non-legacy constructors for tensor deserialization.**
* #18749 Add device and dtype to storage.

Deserialization currently uses legacy constructors.  This is bad because we need to maintain them, but there is a more immediate problem:
1) We are trying to implement device caching on TensorImpl to get rid of a virtual dispatch
2) This doesn't work if one is able to change the device of a Tensor underlying a Variable.
3) Deserialization does 2)

So the plan is to change deserialization, then enforce that we don't change the device out from underneath a Variable.

Differential Revision: D14729513

fbshipit-source-id: 090d6cdb375b94dc1bf4f554b2df243952b8cdc6

5 years agoAdded numpy conversion (#18505)
Iurii Zdebskyi [Wed, 3 Apr 2019 14:22:38 +0000 (07:22 -0700)]
Added numpy conversion (#18505)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18505
ghimport-source-id: f3c9b9251e5793f9e192f587194ddfebb45facc1

Stack from [ghstack](https://github.com/ezyang/ghstack):
* **#18505 [WIP]Added numpy conversion**
* #18166 Bool Tensor for CUDA

Differential Revision: D14646403

fbshipit-source-id: 79d39d692c778ce1981c1d35b1c33e3d93111041

5 years agoRemove THTensor_(newUnfold). (#18773)
Gregory Chanan [Wed, 3 Apr 2019 14:05:16 +0000 (07:05 -0700)]
Remove THTensor_(newUnfold). (#18773)

Summary:
It's not used and unfold's use of `device_guard: False` is scary.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18773

Differential Revision: D14736526

Pulled By: gchanan

fbshipit-source-id: 6281a284bee45fa5038783e4c1ed4d1ed7ca81ab

5 years agotemp fix for flake8 error (#18788)
mingzhe0908 [Wed, 3 Apr 2019 05:49:49 +0000 (22:49 -0700)]
temp fix for flake8 error (#18788)

Summary:
Fix lint error
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18788

Reviewed By: houseroad

Differential Revision: D14741840

Pulled By: mingzhe09088

fbshipit-source-id: 1fa630e3c6e606e3d78fe8293e5b0e7ea1b78da3

5 years agoFix flake8 issues
Igor Fedan [Wed, 3 Apr 2019 04:10:22 +0000 (21:10 -0700)]
Fix flake8 issues

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/18762

Reviewed By: houseroad

Differential Revision: D14734152

Pulled By: ifedan

fbshipit-source-id: 5adf123f88273895ad34ee9041896358d686de08

5 years agoChange ReinitializeTensor to use C10_LOG_FIRST_N (#18531)
Jerry Zhang [Wed, 3 Apr 2019 03:54:28 +0000 (20:54 -0700)]
Change ReinitializeTensor to use C10_LOG_FIRST_N (#18531)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18531

Currently we use C10_LOG_EVERY_MS to log the data type change, but it pollutes the log of some service,
we would like to change it to C10_LOG_FIRST_N to prevent that.

Reviewed By: dzhulgakov

Differential Revision: D14647704

fbshipit-source-id: b84e4002bd4aa94d616133cd1049c3d4ab05386e

5 years agoAdd support for getting TensorProto argument (#18364)
Yinghai Lu [Wed, 3 Apr 2019 03:52:58 +0000 (20:52 -0700)]
Add support for getting TensorProto argument (#18364)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18364

att

Reviewed By: bddppq

Differential Revision: D14584784

fbshipit-source-id: 03f9207d5cf4f7f4b812428a931edbcdcb21ca8d

5 years agomake test module hook use save/load (#18284)
Michael Suo [Wed, 3 Apr 2019 01:06:07 +0000 (18:06 -0700)]
make test module hook use save/load (#18284)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18284
ghimport-source-id: 5a92c03fda19072ffb6afd40e0f56806716c7be6

Stack from [ghstack](https://github.com/ezyang/ghstack):
* #18296 [jit] Add namespacing for ScriptClasses
* **#18284 [jit] make test module hook use save/load**
* #18211 [jit] Turn script_type_parser into a class
* #18148 [jit] python interop for script classes

Instead of python-printing and comparing strings (which does not capture
depdency information, etc.), use save/load on in-memory buffers and
compare the main module contents inside the buffer

Reviewed By: ailzhang

Differential Revision: D14581129

fbshipit-source-id: 52264ae9ce076775ab3fd1a0c32c8d6f6677a903

5 years agoAdd ability to specialize class types to ArgumentSpec (#18314)
Zachary DeVito [Wed, 3 Apr 2019 00:33:06 +0000 (17:33 -0700)]
Add ability to specialize class types to ArgumentSpec (#18314)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18314
ghimport-source-id: 8cecb768d476ab19c9460f39c8f94a764e4cb052

Stack from [ghstack](https://github.com/ezyang/ghstack):
* **#18314 Add ability to specialize class types to ArgumentSpec**
* #18226 Add Slot type to abstract the raw pointers being used for slots.

Differential Revision: D14574395

fbshipit-source-id: cc3af6e56e9ae52990f4a1ad56ecceaa2d493577

5 years agoOperator-level performance microbenchmarks (#18740)
Mingzhe Li [Wed, 3 Apr 2019 00:03:23 +0000 (17:03 -0700)]
Operator-level performance microbenchmarks (#18740)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18740

Test utilities for writing Caffe2/PyTorch performance microbenchmarks. Brief description of the file structure

* benchmark_core.py : core utiltiites for running microbenchmark tests
* benchmark_caffe2.py : Caffe2 specific benchmark utilitites
* benchmark_pytorch.py: PyTorch specific benchmark utilities
* benchmark_runner.py : Main function. Currently it can run the microbenchmark tests in a stand-alone mode. The next step is to have this integrate with AI-PEP.

The utilities are located at https://github.com/pytorch/pytorch/tree/master/test to have access to both Caffe2/PyTorch Python's frontend.

Include two operator microbenchmarks; support both Caffe2/PyTorch:
* MatMul
* Add

Reference: PyTorch benchmarks : https://github.com/pytorch/benchmark/tree/master/timing/python. In this work, we start with two example binary operators MatMul and Add, but eventually we should to cover unary operators like in the PyTorch benchmark repo.

Reviewed By: zheng-xq

Differential Revision: D13887111

fbshipit-source-id: b7a56b95448c9ec3e674b0de0ffb96af4439bfce

5 years agoBool Tensor for CUDA (#18166)
Iurii Zdebskyi [Tue, 2 Apr 2019 23:10:43 +0000 (16:10 -0700)]
Bool Tensor for CUDA (#18166)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18166
ghimport-source-id: a8e2ba2d966e49747a55701c4f6863c5e24d6f14

Stack from [ghstack](https://github.com/ezyang/ghstack):
* **#18166 Bool Tensor for CUDA**
* #18165 Resolved comments from Bool Tensor for CPU PR
------

This PR enables bool tensor creation and some basic operations for the CPU backend. This is a part of Bool Tensor feature implementation work. The whole plan looks like this:
1. Storage Implementation [Done]
2. Tensor Creation.
a) CPU [Done]
b) CUDA [This PR]
3. Tensor Conversions.
4. Tensor Indexing.
5. Tensor Operations.
6. Back compatibility related changes.

Change:
Enable bool tensor in CUDA with the following operations:

    torch.zeros
    torch.tensor
    torch.ones
    torch.rand/rand_like/randint/randint_like
    torch.full
    torch.full_like
    torch.empty
    torch.empty_like

Tested via unit tests and local scripts.

Differential Revision: D14605104

fbshipit-source-id: b7d7340a7d70edd03a109222d271e68becba762c

5 years agoAdd helpful information to the gradient/inplace operation exception (#18523)
Jan Schlüter [Tue, 2 Apr 2019 22:15:31 +0000 (15:15 -0700)]
Add helpful information to the gradient/inplace operation exception (#18523)

Summary:
To debug a `one of the variables needed for gradient computation has been modified by an inplace operation` error, I wanted to know *which* variable has been modified, so I extended the error message with what information is easily available at this point.

Before:
```
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation
```

After:
```
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [80, 1]], which is output 0 of UnsqueezeBackward0, is at version 1, not expected version 0. Hint: enable anomaly detection to find the forward pass operation which modified it.
```

The hint to enable anomaly detection is only shown when it is not enabled. It's meant to save people some googling. I'd even go further and reference `torch.autograd.set_detect_anomaly(True)`, but maybe we're not running Python?

Disclaimer: I haven't looked at other parts of the code to check if using `std::stringstream` is acceptable practice, let me know if it isn't. Similarly, I haven't checked about indentation practices.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18523

Differential Revision: D14683249

Pulled By: soumith

fbshipit-source-id: f97a99d4aabea7461df766d66cd72300b48e2350

5 years agobuild_variables.py: turn on link_whole for _C_impl library. (#18763)
Mikhail Zolotukhin [Tue, 2 Apr 2019 22:14:18 +0000 (15:14 -0700)]
build_variables.py: turn on link_whole for _C_impl library. (#18763)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18763

Without `link_whole` flag in opt-builds some of the files are not linked into `_C_impl` library, which causes some of static initializers not to run (namely, registering an cutomPythonOperation from python_interpreter.cpp). This diff fixes it.

Differential Revision: D14732471

fbshipit-source-id: 57cff6b4b6d479ad7ab7fd29f677746d91d6ff45

5 years agoFix windows msbuild bug (#18748)
vaeksare [Tue, 2 Apr 2019 21:25:28 +0000 (14:25 -0700)]
Fix windows msbuild bug (#18748)

Summary:
Fix the bug introduced by #18681 where an undefined variable was being used to limit max cpu count when building for Windows without Ninja.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18748

Differential Revision: D14733209

Pulled By: soumith

fbshipit-source-id: 52fc0dd4dde99da75a6956b63f02da2e647eed4f

5 years agotorch.cross' dim default changed to c10::optional instead of int=-1 (#17582)
Igor Fedan [Tue, 2 Apr 2019 20:18:20 +0000 (13:18 -0700)]
torch.cross' dim default changed to c10::optional instead of int=-1 (#17582)

Summary:
Argument dim=-1 doesn't work for torch.cross. The signature of the torch.cross has been changed to c10::optional<int64_t> dim instead of int64_t. So based on document "If dim is not given, it defaults to the first dimension found with the size 3." and if dim is specified (even negative) it will use the correspondent dim.

Fixes #17229
Pull Request resolved: https://github.com/pytorch/pytorch/pull/17582

Differential Revision: D14483063

Pulled By: ifedan

fbshipit-source-id: f9699093ec401cb185fd33ca4563c8a46cdcd746

5 years agoFix multi-configuration on Windows CMake (CUDA) (#18548)
Sacha [Tue, 2 Apr 2019 20:15:10 +0000 (13:15 -0700)]
Fix multi-configuration on Windows CMake (CUDA) (#18548)

Summary:
Multiple configurations is the default (eg. Release;Debug) on Windows and this check always broke this configuration as CMAKE_BUILD_TYPE was not set. The workaround was to always set CMAKE_BUILD_TYPE to Debug or Release, which was very unfortunate.

The correct method is to use generator expressions that expand depending on the current CONFIG being processed.

Side note: Anywhere else CMAKE_BUILD_TYPE is checked should probably be fixed too.
Note that the CMakeLists.txt forces it in to Release mode. However, I came across this error when importing the prebuilt Config in to another project, where CMAKE_BUILD_TYPE was not set.

> 3>CMake Error at pre_built/pytorch-1.0.1/share/cmake/Caffe2/public/cuda.cmake:380 (message):
> 3>  Unknown cmake build type:

Proper support for configurations would mean we can build debug and release at the same time and as you can see, it is less CMake code.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18548

Differential Revision: D14730790

Pulled By: ezyang

fbshipit-source-id: 70ae16832870d742c577c34a50ec7564c3da0afb

5 years agoFix flake8 issues in gragrad test
Igor Fedan [Tue, 2 Apr 2019 19:32:52 +0000 (12:32 -0700)]
Fix flake8 issues in gragrad test

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/18727

Differential Revision: D14724887

Pulled By: ifedan

fbshipit-source-id: 8c1db6460303e746e4aea0142302b8d61277c067

5 years agoRegister operators by passing arguments to RegisterOperators constructor (#18577)
Sebastian Messmer [Tue, 2 Apr 2019 19:23:15 +0000 (12:23 -0700)]
Register operators by passing arguments to RegisterOperators constructor (#18577)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18577

This is also part of the legacy API and we need to support it if we want to replace it.

Reviewed By: dzhulgakov

Differential Revision: D14671432

fbshipit-source-id: 007abf4ab816647a509fc08e35d79b6c1aa55b03

5 years agoAllow registering an operator schema without a kernel (#18551)
Sebastian Messmer [Tue, 2 Apr 2019 19:23:13 +0000 (12:23 -0700)]
Allow registering an operator schema without a kernel (#18551)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18551

This is helpful for defining a set of operators as an interface but not adding concrete kernels just yet.
The registration logic will ensure that any other libraries that add kernels for these schemas exactly match the schema defined here.

Reviewed By: dzhulgakov

Differential Revision: D14660208

fbshipit-source-id: 7adb5a4876cff5a0ad21d92d8c450cb889f00cc3

5 years agoImprove compiler error messages of the op registration API (#18550)
Sebastian Messmer [Tue, 2 Apr 2019 19:23:13 +0000 (12:23 -0700)]
Improve compiler error messages of the op registration API (#18550)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18550

When the operator registration API is used wrongly, in most cases we should now get a nice compiler error
instead of weird template error messages.

This is done by making the enable_if conditions more broad so they also match error cases,
but then having static_asserts against these error cases inside the function.
Before that, since the function didn't match, the error message said something like "no function found to match your call",
now it will show the error message specified in the static_asserts.

Reviewed By: dzhulgakov

Differential Revision: D14659178

fbshipit-source-id: 7ca4fb72d9051eadf0a7e2717b962bf1213a52b2

5 years agoImprove and test error messages for signature mismatches (#18547)
Sebastian Messmer [Tue, 2 Apr 2019 19:23:13 +0000 (12:23 -0700)]
Improve and test error messages for signature mismatches (#18547)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18547

- Argument indices in the error messages are 1-indexed not 0-indexed.
- Add test cases that a mismatching signature actually shows the correct error messages

Reviewed By: dzhulgakov

Differential Revision: D14656695

fbshipit-source-id: 55e45634baa3117e18b8687ea6b2a2f83715bdf6

5 years agoEnable gmock and fix system gtest issue (#18706)
Sebastian Messmer [Tue, 2 Apr 2019 19:23:13 +0000 (12:23 -0700)]
Enable gmock and fix system gtest issue (#18706)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18706

- Enable gmock
- Fix issue where the gtest source files in third_party would include system gtest headers

Reviewed By: ezyang

Differential Revision: D14715302

fbshipit-source-id: 5335390913e651bda85c69d7ea9b5c1bce58f172

5 years agoEmergency workaround for apt-get failure. (#18733)
Edward Yang [Tue, 2 Apr 2019 17:46:16 +0000 (10:46 -0700)]
Emergency workaround for apt-get failure. (#18733)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18733
ghimport-source-id: b56766fb4b1084d8a7947cf622275d44e325141b

Stack from [ghstack](https://github.com/ezyang/ghstack):
* **#18733 Emergency workaround for apt-get failure.**

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

Differential Revision: D14725779

fbshipit-source-id: 6855347853a3f13461ca267ed563e2db5815166e

5 years agoFix clang-tidy errors in torch/csrc/distributed
Pieter Noordhuis [Tue, 2 Apr 2019 17:29:54 +0000 (10:29 -0700)]
Fix clang-tidy errors in torch/csrc/distributed

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/18709

Differential Revision: D14725936

Pulled By: pietern

fbshipit-source-id: 307bc446d53da5d0e04d730bb51b7fb29212ace3

5 years agoUndefined behavior with memset of std::string to 0 (#18703)
Eli Amesefe [Tue, 2 Apr 2019 17:07:22 +0000 (10:07 -0700)]
Undefined behavior with memset of std::string to 0 (#18703)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18703

 `zeroPtr` is sometimes a `std::string` tensor, so `memset` to 0 is undefined behavior.

This might be accidentally safe with `std::string` implementation that use SSO (Small String Optimization), but will crash otherwise.

Reviewed By: zheng-xq

Differential Revision: D14714458

fbshipit-source-id: 012a18464e6514d38ff791509b88ddc3fc55b2b1

5 years agoRevert D14717015: [pytorch][PR] fix nccl compilation to make sure it compiles for...
Soumith Chintala [Tue, 2 Apr 2019 16:36:05 +0000 (09:36 -0700)]
Revert D14717015: [pytorch][PR] fix nccl compilation to make sure it compiles for architectures that pytorch compiles for

Differential Revision:
D14717015

Original commit changeset: 4aac036f57e5

fbshipit-source-id: c820b8dfb27564271e6b80e133fe655658a7c25c

5 years agoAutomatic update of fbcode/onnx to f0d7df2c643c4e37f1fd7735ef02c972c4d19fb5 (#18695)
Lu Fang [Tue, 2 Apr 2019 16:11:14 +0000 (09:11 -0700)]
update of fbcode/onnx to f0d7df2c643c4e37f1fd7735ef02c972c4d19fb5 (#18695)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18695

Previous import was fb1a80692c1ab0bd27b1072f2e7bffacba336777

Included changes:
- **[f0d7df2c](https://github.com/onnx/onnx/commit/f0d7df2c)**: fix testcase names of maxpool_2d_ceil and averagepool_2d_ceil (#1896) <karljang>

Reviewed By: zrphercule

Differential Revision: D14709993

fbshipit-source-id: 7fe2145a481ea2c1b6d85ba1c85c662200a53241

5 years agoAdding pin_memory kwarg to zeros, ones, empty, ... tensor constructors. (#18455)
Vitaly Fedyunin [Tue, 2 Apr 2019 15:44:27 +0000 (08:44 -0700)]
Adding pin_memory kwarg to zeros, ones, empty, ... tensor constructors. (#18455)

Summary:
Make it possible to construct a pinned memory tensor without creating a storage first and without calling pin_memory() function. It is also faster, as copy operation is unnecessary.

Supported functions:
```python
torch.rand_like(t, pin_memory=True)
torch.randn_like(t, pin_memory=True)
torch.empty_like(t, pin_memory=True)
torch.full_like(t, 4, pin_memory=True)
torch.zeros_like(t, pin_memory=True)
torch.ones_like(t, pin_memory=True)
torch.tensor([10,11], pin_memory=True)
torch.randn(3, 5, pin_memory=True)
torch.rand(3, pin_memory=True)
torch.zeros(3, pin_memory=True)
torch.randperm(3, pin_memory=True)
torch.empty(6, pin_memory=True)
torch.ones(6, pin_memory=True)
torch.eye(6, pin_memory=True)
torch.arange(3, 5, pin_memory=True)
```

Part of the bigger: `Remove Storage` plan.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18455

Reviewed By: ezyang

Differential Revision: D14672084

Pulled By: VitalyFedyunin

fbshipit-source-id: 9d0997ec00f59500ee018f8b851934d334012124

5 years agoImprove Backend comment. (#18567)
Edward Yang [Tue, 2 Apr 2019 15:03:29 +0000 (08:03 -0700)]
Improve Backend comment. (#18567)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18567
ghimport-source-id: 1e50e611a3afcfae86828b7afe06c3fdc6a7bef7

Stack from [ghstack](https://github.com/ezyang/ghstack):
* **#18567 Improve Backend comment.**

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

Differential Revision: D14666189

fbshipit-source-id: 64a41c4a998b1a59ff780d1ae06fa16e5ef3c7c4

5 years agoExpose alias multinomial methods to ATen (#17904)
vishwakftw [Tue, 2 Apr 2019 14:53:34 +0000 (07:53 -0700)]
Expose alias multinomial methods to ATen (#17904)

Summary:
This PR exposes the multinomialAliasSetup and multinomialAliasDraw methods.

cc: neerajprad
Pull Request resolved: https://github.com/pytorch/pytorch/pull/17904

Differential Revision: D14700205

Pulled By: ezyang

fbshipit-source-id: 16462fb1f1ef1d560fd586632ea356b23e966ee3

5 years agoUpdate cpp_extension.py (#18638)
BloodAxe [Tue, 2 Apr 2019 14:49:40 +0000 (07:49 -0700)]
Update cpp_extension.py (#18638)

Summary:
Hi. It seems that when building CPP-extensions with CUDA for Windows, an `extra_cuda_cflags` options are not properly forwarded to `nvcc`.

Use of extra CUDA options is necessary to build, for instance, a InplaceABN (https://github.com/mapillary/inplace_abn), which requires `--expt-extended-lambda` option.

This PR adds one line that correctly appends `extra_cuda_cflags`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18638

Differential Revision: D14704270

Pulled By: ezyang

fbshipit-source-id: e1e330d193d9afd5707a5437a74c0499460d2b90

5 years agofix typo
Mark Pare [Tue, 2 Apr 2019 14:48:45 +0000 (07:48 -0700)]
fix typo

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/18653

Differential Revision: D14713920

Pulled By: ezyang

fbshipit-source-id: 170295a162dd23916c1dcc9330918d33277cc9ed

5 years agoKill LegacyBridge functions that don't do multiple dispatch. (#18696)
Gregory Chanan [Tue, 2 Apr 2019 14:31:54 +0000 (07:31 -0700)]
Kill LegacyBridge functions that don't do multiple dispatch. (#18696)

Summary:
At some point, we needed these functions to deal with autograd dispatching to the sparse of TH version of a backwards.  But we rewrote all backwards definitions in terms of native functions, so this is no longer necessary.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18696

Differential Revision: D14710834

Pulled By: gchanan

fbshipit-source-id: b22568c58eefc79d672555bd8832398ccd965cb7

5 years agoUpdating submodules
svcscm [Tue, 2 Apr 2019 13:47:23 +0000 (06:47 -0700)]
Updating submodules

Reviewed By: zpao

fbshipit-source-id: da3cd711bb81b07c6c284426ffc5e10a969b0d2b

5 years agoadd Int8FCRelu (#18673)
Jongsoo Park [Tue, 2 Apr 2019 06:45:01 +0000 (23:45 -0700)]
add Int8FCRelu (#18673)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18673

Add a fused FC + Relu

Reviewed By: csummersea

Differential Revision: D14667055

fbshipit-source-id: d88fefba008fc0ca450291532d2b320694c6b785

5 years agoFix uninitialized value in pickler (#18678)
David Riazati [Tue, 2 Apr 2019 00:31:53 +0000 (17:31 -0700)]
Fix uninitialized value in pickler (#18678)

Summary:
Fixes #18671
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18678

Differential Revision: D14708969

Pulled By: driazati

fbshipit-source-id: d372c6e3a2a3d3fc48d8afc1fa6807f2ce0e5c6e

5 years agofixes multiprocessing serialization for integer nn.Parameter (#18639)
Soumith Chintala [Tue, 2 Apr 2019 00:07:27 +0000 (17:07 -0700)]
fixes multiprocessing serialization for integer nn.Parameter (#18639)

Summary:
Fixes https://github.com/pytorch/pytorch/issues/17345
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18639

Differential Revision: D14711565

Pulled By: soumith

fbshipit-source-id: 0063ed138a215b95d6571dcd68b18569714abe19

5 years agofix nccl compilation to make sure it compiles for architectures that pytorch compiles...
Soumith Chintala [Tue, 2 Apr 2019 00:07:08 +0000 (17:07 -0700)]
fix nccl compilation to make sure it compiles for architectures that pytorch compiles for (#18704)

Summary:
cc: t-vi gchanan zou3519

This fixes https://github.com/pytorch/pytorch/issues/18359
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18704

Differential Revision: D14717015

Pulled By: soumith

fbshipit-source-id: 4aac036f57e564b05d759662e8ad7a80170901c0

5 years agoMore type stubs (#18511)
Jon Malmaud [Mon, 1 Apr 2019 22:56:21 +0000 (15:56 -0700)]
More type stubs (#18511)

Summary:
Added stubs for:

* The `device` module
* The `cuda` module
* Parts of the `optim` module
* Began adding stubs for the `autograd` module. I'll annotate more later but `no_grad` and friends are probably the most used exports from it so it seemed like a good place to start.

This would close #16996, although comments on that issue reference other missing stubs so maybe it's worth keeping open as an umbrella issue.

The big remaining missing package is `nn`.

Also added a `py.typed` file so mypy will pick up on the type stubs. That closes #17639.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18511

Differential Revision: D14715053

Pulled By: ezyang

fbshipit-source-id: 9e4882ac997063650e6ce47604b3eaf1232c61c9

5 years agoNCCL build fix WITH_DISTRIBUTED=1.
Gregory Chanan [Mon, 1 Apr 2019 22:51:44 +0000 (15:51 -0700)]
NCCL build fix WITH_DISTRIBUTED=1.

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/18691

Reviewed By: ezyang

Differential Revision: D14706205

Pulled By: gchanan

fbshipit-source-id: 802f19bfd7df3703c0dbce03036e2f2e32eb3efb

5 years agocaffe2 - set up correct inheritance structure for remaining operator test classes...
Duc Ngo [Mon, 1 Apr 2019 22:49:56 +0000 (15:49 -0700)]
caffe2 - set up correct inheritance structure for remaining operator test classes (#18622)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18622

Set up correct inheritance structure for remaining operator test classes

Reviewed By: ezyang

Differential Revision: D14685941

fbshipit-source-id: a6b1b3be325935b7fec7515be13a4994b3016bf0

5 years agoPeephole Optimize Shape Ops (#18549)
Elias Ellison [Mon, 1 Apr 2019 22:33:35 +0000 (15:33 -0700)]
Peephole Optimize Shape Ops (#18549)

Summary:
Peephole optimize ops that just require Dimensioned Tensor Type, which is what we specialize graphs on.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18549

Differential Revision: D14690827

Pulled By: eellison

fbshipit-source-id: 9d7439eb584f0a5b877f5aa53cf80150f00e7e5f

5 years agoDeprecated lambda based API (#18542)
Sebastian Messmer [Mon, 1 Apr 2019 21:53:08 +0000 (14:53 -0700)]
Deprecated lambda based API (#18542)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18542

This adds the deprecated API for defining kernels as lambdas. The new API for defining kernels as lambdas was introduced in D14653005.

Reviewed By: dzhulgakov

Differential Revision: D14653551

fbshipit-source-id: 99900f1436716c69e52c83b68333b642ec2c8558

5 years agodeprecated function based API (#18444)
Sebastian Messmer [Mon, 1 Apr 2019 21:53:08 +0000 (14:53 -0700)]
deprecated function based API (#18444)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18444

This adds the deprecated function based API to c10::RegisterOperators().
This is the API currently exposed under jit::RegisterOperators() and we need to support it for backwards compatibility.

Reviewed By: dzhulgakov

Differential Revision: D14514218

fbshipit-source-id: c77676851cfd431d66f18fd8038cf153a3a7d7cc

5 years agoRevert "Tensor construction codemod(raw_mutable_data) (#16373)" (#18680)
Junjie Bai [Mon, 1 Apr 2019 21:30:09 +0000 (14:30 -0700)]
Revert "Tensor construction codemod(raw_mutable_data) (#16373)" (#18680)

Summary:
This reverts commit d73c830e236f5b980e5c91914b818d150b60278c.

We have observed significant perf drop when training ResNext101 with multiple amd GPUs:

Before:
https://ci.pytorch.org/jenkins/job/caffe2-builds/job/py2-clang7-rocmdeb-ubuntu16.04-bench/1636/console
2 GPUs ResNext training got 150\~160 imgs/sec
4 GPUs ResNext training got 270\~280 imgs/sec

After:
https://ci.pytorch.org/jenkins/job/caffe2-builds/job/py2-clang7-rocmdeb-ubuntu16.04-bench/1637/console
Both 2 and 4 GPUs ResNext training drop to 110\~120 imgs/sec

Similar perf drop are seen on ResNet50 training jobs as well.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18680

Differential Revision: D14702941

Pulled By: bddppq

fbshipit-source-id: 828141805afc23f25c08d4a2eb6d4b99f817c128

5 years agoC++ handler for gradient reduction (#18251)
Pieter Noordhuis [Mon, 1 Apr 2019 21:27:03 +0000 (14:27 -0700)]
C++ handler for gradient reduction (#18251)

Summary:
This commit adds the `c10d::Reducer` class that hooks into autograd
and performs gradient bucketing and reduction. These are the core
parts of `nn.parallel.DistributedDataParallel` that up to now were
only usable for CUDA models.

This should enable the following:

* Distributed data parallelism for models defined using the C++ frontend.
* Allow overlap of gradient computation and reduction for non-CUDA models.
* Enable distributed data parallelism for models with some unused parameters.

This does not include any logic for computing bucket assignment, which
can be done separately; either by observing autograd execution order
(this is what Apex does), or by assigning buckets based on some
maximum byte size, or both.

Also see #17757 and #13273.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18251

Reviewed By: mrshenli

Differential Revision: D14571899

Pulled By: pietern

fbshipit-source-id: 20f95eefd288dfe8cfffe0a28ca22fa7c9c3cd4c

5 years agoUpdating submodules
svcscm [Mon, 1 Apr 2019 20:54:40 +0000 (13:54 -0700)]
Updating submodules

Reviewed By: zpao

fbshipit-source-id: 735fc388bff7066e8f46526266a73bf35e121442

5 years agoadd ConvRelu schema (#18693)
Jongsoo Park [Mon, 1 Apr 2019 20:02:02 +0000 (13:02 -0700)]
add ConvRelu schema (#18693)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18693

As title

Reviewed By: protonu

Differential Revision: D14662880

fbshipit-source-id: 3664faa660a04e1f528a413d2a1700b872c3c684

5 years agooffload scripts from win-test.sh
Karl Ostmo [Mon, 1 Apr 2019 20:01:29 +0000 (13:01 -0700)]
offload scripts from win-test.sh

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/18601

Differential Revision: D14711856

Pulled By: kostmo

fbshipit-source-id: 75fe620541fe2903f69a53dbd1b6d51a0d718113

5 years agoSome fixes for the build script on Windows (#18681)
peter [Mon, 1 Apr 2019 19:37:00 +0000 (12:37 -0700)]
Some fixes for the build script on Windows (#18681)

Summary:
Fixes https://discuss.pytorch.org/t/pytorch-build-from-source-on-windows/40288/13?u=peterjc123.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18681

Differential Revision: D14711039

Pulled By: soumith

fbshipit-source-id: f7e1a94b163064c055670b2925cd4502e7773599

5 years agoFix for double backwards tests (#18190)
Igor Fedan [Mon, 1 Apr 2019 19:30:29 +0000 (12:30 -0700)]
Fix for double backwards tests (#18190)

Summary:
If none of the outputs require_grad, we don't actually check gradgrad, instead we will check that their numerical gradients are 0.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18190

Differential Revision: D14563388

Pulled By: ifedan

fbshipit-source-id: a4eb94c9eb60f14dbe6986cd8cef1fe78a7bc839

5 years agoAdd string index/slice operations (#18247)
David Riazati [Mon, 1 Apr 2019 18:58:28 +0000 (11:58 -0700)]
Add string index/slice operations (#18247)

Summary:
Adds support for string indexing (`"a"[0]`) and slicing (`"abc"[1:3]`)
to script.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18247

Differential Revision: D14574486

Pulled By: driazati

fbshipit-source-id: 4b42aa0881e5398ea7f112be46c0335e6e19dced

5 years agoRe-land Parsing file check (#18570)
eellison [Mon, 1 Apr 2019 18:48:39 +0000 (11:48 -0700)]
Re-land Parsing file check (#18570)

Summary:
The last time I tried to land it there was a merge race with the docs coverage test lol. Re-landing with the fix.

Re-land of https://github.com/pytorch/pytorch/pull/18304
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18570

Reviewed By: driazati

Differential Revision: D14707285

Pulled By: eellison

fbshipit-source-id: 3a0265928aa8cad78961723d8bf0fbf871fdb71d

5 years agoCreate Node2Vec ModuleKeeper
Ru Li [Mon, 1 Apr 2019 17:31:08 +0000 (10:31 -0700)]
Create Node2Vec ModuleKeeper

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/18504

Reviewed By: sunnieshang

Differential Revision: D14632091

fbshipit-source-id: d4544866552dc6bcbc7515be9e88cb11e7622a44

5 years agouse acc16 only when n>128 and k>128 in Skylake (#18672)
Jongsoo Park [Mon, 1 Apr 2019 15:49:37 +0000 (08:49 -0700)]
use acc16 only when n>128 and k>128 in Skylake (#18672)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18672

In Skylake, when n < 128 or k < 128, acc16 is slower.

Reviewed By: jianyuh

Differential Revision: D14700576

fbshipit-source-id: 80ca9f1af4626637eed9c5ca49f95ae744811189

5 years agoMove ideep singleton registration to ATen from C2. (#18335)
Gregory Chanan [Mon, 1 Apr 2019 14:57:48 +0000 (07:57 -0700)]
Move ideep singleton registration to ATen from C2. (#18335)

Summary:
Since we are going to add ideep to ATen, and ATen is always compiled, it makes sense to have the registration in ATen rather than C2.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18335

Reviewed By: bddppq

Differential Revision: D14578652

Pulled By: gchanan

fbshipit-source-id: 4d77fcfc21a362b21d5291a127498aa722548873

5 years agoCreate torch/lib directory before copying _C.lib on Windows environment. (#18666)
Shuichi KITAGUCHI [Mon, 1 Apr 2019 14:24:27 +0000 (07:24 -0700)]
Create torch/lib directory before copying _C.lib on Windows environment. (#18666)

Summary:
`python setup.py develop` fails with following messages.
~~~
...
-- Building with NumPy bindings
-- Not using cuDNN
-- Not using MIOpen
-- Not using CUDA
-- Using MKLDNN
-- Not using NCCL
-- Building without distributed package

Copying extension caffe2.python.caffe2_pybind11_state
Copying caffe2.python.caffe2_pybind11_state from torch\Lib\site-packages\caffe2\python\caffe2_pybind11_state.cp37-win_amd64.pyd to C:\data\source\pytorch\build\lib.win-amd64-3.7\caffe2\python\caffe2_pybind11_state.cp37-win_amd64.pyd
copying torch\Lib\site-packages\caffe2\python\caffe2_pybind11_state.cp37-win_amd64.pyd -> C:\data\source\pytorch\build\lib.win-amd64-3.7\caffe2\python
building 'torch._C' extension
creating build\temp.win-amd64-3.7
creating build\temp.win-amd64-3.7\Release
creating build\temp.win-amd64-3.7\Release\torch
creating build\temp.win-amd64-3.7\Release\torch\csrc
...
creating C:\data\source\pytorch\build\lib.win-amd64-3.7\torch
C:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.16.27023\bin\HostX64\x64\link.exe /nologo /INCREMENTAL:NO /LTCG /nodefaultlib:libucrt.lib ucrt.lib /DLL /MANIFEST:EMBED,ID=2 /MANIFESTUAC:NO /LIBPATH:C:\data\source\pytorch\torch\lib /LIBPATH:C:\data\dlenv\libs /LIBPATH:C:\data\dlenv\PCbuild\amd64 "/LIBPATH:C:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.16.27023\ATLMFC\lib\x64" "/LIBPATH:C:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.16.27023\lib\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\NETFXSDK\4.6.1\lib\um\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\10\lib\10.0.17763.0\ucrt\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\10\lib\10.0.17763.0\um\x64" shm.lib torch_python.lib /EXPORT:PyInit__C build\temp.win-amd64-3.7\Release\torch/csrc/stub.obj /OUT:build\lib.win-amd64-3.7\torch\_C.cp37-win_amd64.pyd /IMPLIB:build\temp.win-amd64-3.7\Release\torch/csrc\_C.cp37-win_amd64.lib /NODEFAULTLIB:LIBCMT.LIB
   ライブラリ build\temp.win-amd64-3.7\Release\torch/csrc\_C.cp37-win_amd64.lib とオブジェクト build\temp.win-amd64-3.7\Release\torch/csrc\_C.cp37-win_amd64.exp を作成中
コード生成しています。
コード生成が終了しました。
copying build\lib.win-amd64-3.7\torch\_C.cp37-win_amd64.pyd -> torch
copying build\lib.win-amd64-3.7\caffe2\python\caffe2_pybind11_state.cp37-win_amd64.pyd -> caffe2\python
copying build/temp.win-amd64-3.7/Release/torch/csrc/_C.cp37-win_amd64.lib -> build/lib.win-amd64-3.7/torch/lib/_C.lib
error: could not create 'build/lib.win-amd64-3.7/torch/lib/_C.lib': No such file or directory
~~~

When `python setup.py install` is executed, `torch/lib`  has been created by previous process (copying many files) and this copy succeeds. But in develop mode, that process does not executed and this copy fails.

This patch creates `torch/lib` directory if do not exist.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18666

Differential Revision: D14704269

Pulled By: ezyang

fbshipit-source-id: b2d7c698a906b945bf34bb78f17b91b4fdfd3294

5 years agoMove flags that do not work on MSVC (#18686)
Sacha [Mon, 1 Apr 2019 14:23:06 +0000 (07:23 -0700)]
Move flags that do not work on MSVC (#18686)

Summary:
MSVC errors on these flags as they are not supported
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18686

Differential Revision: D14704254

Pulled By: ezyang

fbshipit-source-id: 936d33ed6b7474d7774a49505cdac50dbe8dd99a

5 years agoFix unused lambda capture warnings (#18662)
Junjie Bai [Mon, 1 Apr 2019 05:26:27 +0000 (22:26 -0700)]
Fix unused lambda capture warnings (#18662)

Summary:
```
aten/src/ATen/native/cpu/DistanceOpsKernel.cpp.DEFAULT.cpp:109:104: warning: lambda capture 'combs' is not used [-Wunused-lambda-capture]
    parallel_for(0, combs, internal::GRAIN_SIZE / (16 * m), [p, self_start, self_end, n, m, res_start, combs](int64_t k, int64_t end) {
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18662

Differential Revision: D14699379

Pulled By: bddppq

fbshipit-source-id: 5062d4327bb5f7b485c2ffa30c98e10576416f03

5 years agohandle a rare case of histogram min is inf/nan (#18239)
Jongsoo Park [Mon, 1 Apr 2019 04:25:17 +0000 (21:25 -0700)]
handle a rare case of histogram min is inf/nan (#18239)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18239

When min is inf or nan, we get UBSAN errors

Reviewed By: csummersea

Differential Revision: D14537668

fbshipit-source-id: e70ffb5ecd2b10793356070c69fdabf8f25b203e

5 years agoDelete duplicated technical content from contribution_guide.rst (#18628)
Edward Yang [Mon, 1 Apr 2019 02:08:03 +0000 (19:08 -0700)]
Delete duplicated technical content from contribution_guide.rst (#18628)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18628
ghimport-source-id: d94b81a6f303883d97beaae25344fd591e13ce52

Stack from [ghstack](https://github.com/ezyang/ghstack):
* #18629 Provide flake8 install instructions.
* **#18628 Delete duplicated technical content from contribution_guide.rst**

There's useful guide in contributing_guide.rst, but the
technical bits were straight up copy-pasted from CONTRIBUTING.md,
and I don't think it makes sense to break the CONTRIBUTING.md
link.  Instead, I deleted the duplicate bits and added a cross
reference to the rst document.

Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Differential Revision: D14701003

fbshipit-source-id: 3bbb102fae225cbda27628a59138bba769bfa288

5 years agoProvide flake8 install instructions. (#18629)
Edward Yang [Mon, 1 Apr 2019 01:56:12 +0000 (18:56 -0700)]
Provide flake8 install instructions. (#18629)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18629
ghimport-source-id: 66a8871c56ffcfa7d4bfdf601e180fae99194e28

Stack from [ghstack](https://github.com/ezyang/ghstack):
* **#18629 Provide flake8 install instructions.**
* #18628 Delete duplicated technical content from contribution_guide.rst

Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Differential Revision: D14701004

fbshipit-source-id: b64292f0ef01b7894cf6b9ff8d5fd9e921c8d162

5 years agoAdding quantized tensor shape/type info support for caffe2=>glow in caffe2 side ...
Rui Zhu [Mon, 1 Apr 2019 00:37:17 +0000 (17:37 -0700)]
Adding quantized tensor shape/type info support for caffe2=>glow in caffe2 side (#18621)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18621

This diff added caffe2 support for onnxifi quantization.

Reviewed By: yinghai

Differential Revision: D14648767

fbshipit-source-id: 4ddb492cacbba6142305866e6dbb875880acaea3

5 years agoFix test on windows (#18667)
David Riazati [Sun, 31 Mar 2019 23:17:28 +0000 (16:17 -0700)]
Fix test on windows (#18667)

Summary:
Breakage in #18188
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18667

Differential Revision: D14700133

Pulled By: driazati

fbshipit-source-id: 4cc26bd579fc1b074b3bef6046cc1030facee130

5 years agoEnforce check ad in test_jit (#18509)
Ailing Zhang [Sun, 31 Mar 2019 15:41:46 +0000 (08:41 -0700)]
Enforce check ad in test_jit (#18509)

Summary:
If a test triggers autodiff, it must have a `DifferentiableGraph` in its differentiated forward graph, and this subgraph must have either the original aten node, or the corresponding nodes used in AD formula.

Typically a forward differentiable graph looks like this:
```
graph(%i0 : Float(),
      %i1 : Float()):
  %3 : Float() = prim::DifferentiableGraph_0(%i0, %i1)
  return (%3)
with prim::DifferentiableGraph_0 = graph(%0 : Float(),
      %1 : Float()):
  %2 : Float() = aten::max(%0, %1)
  return (%2)
```
which tells us `aten::max(Tensor self, Tensor other) -> Tensor` is symbolically differentiable.

Update: there're a lot of cases (fusions/ConstantChunk/python implementations) that breaks it so I decided to make the check optionally take node names if different from function name.
~~[OLD]Theoretically I could also check if `aten::max` is in the differentiable block or not to be more precise, but there're also cases like `chunk` where in a differentiable block it's replaced with a prim node (ConstantChunk) and we will have to special case them. Any suggestions here (to be more precise or no) is very welcome!~~

We used to have a list containing nn tests should be run against AD, I moved it to an field when constructing our test(either torch or nn). I think it's cleaner this way, and it matches the fact that for the same op we support one schema of it but not all, in this way we could just turn on the corresponding test which triggers that supported schema.

cc: apaszke zdevito wanchaol ngimel for a review

[Done] :
- Going through a manual second pass of all tests to check if they should enable AD test or not....
- Add a readme about how to add AD for an op and how to add/enable its test in test_jit.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18509

Differential Revision: D14696811

Pulled By: ailzhang

fbshipit-source-id: c5e693277baac585cd3aed5ab2c0e7faa5e6f29f

5 years agoUse proper isnan check
Junjie Bai [Sun, 31 Mar 2019 09:05:14 +0000 (02:05 -0700)]
Use proper isnan check

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/18663

Differential Revision: D14699385

Pulled By: bddppq

fbshipit-source-id: 596ad3371e7704802591e49f7e1c55dc6cd2896f

5 years agopad_circular -> _pad_circular (#18608)
Soumith Chintala [Sat, 30 Mar 2019 20:24:11 +0000 (13:24 -0700)]
pad_circular -> _pad_circular (#18608)

Summary:
pad_circular is really private, as circular padding is exposed via `F.pad`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18608

Differential Revision: D14691704

Pulled By: soumith

fbshipit-source-id: 8c2f90596feed670976115041efed3ca071e8306

5 years agoFix wrap(at::Scalar) (#18632)
Will Feng [Sat, 30 Mar 2019 18:28:44 +0000 (11:28 -0700)]
Fix wrap(at::Scalar) (#18632)

Summary:
Problem:
```cpp
// This function expects a `Variable` as input
inline PyObject* wrap(at::Tensor tensor) {
  return THPVariable_Wrap(Variable(std::move(tensor)));
}

inline PyObject* wrap(at::Scalar scalar) {
  // This function calls `wrap(at::Tensor tensor)` (the function above), but since
  // `scalar_to_tensor(...)` returns a `Tensor` and not a `Variable`, the call to
  // `wrap(at::Tensor tensor)` will fail with "Tensor that was converted to Variable
  // was not actually a Variable", which is not what we want.
  return wrap(scalar_to_tensor(scalar));
}
```

The right fix is to call `make_variable(...)` with the tensor returned from `scalar_to_tensor(scalar)`.

This unblocks https://github.com/pytorch/pytorch/pull/18230 as it is the only patch that hits this code path now. All other native functions that return Scalar (such as `item()` or `_local_scalar_dense()`) either has custom-defined implementation that doesn't go through this path, or is not exposed to Python at all.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18632

Differential Revision: D14689293

Pulled By: yf225

fbshipit-source-id: be7ba5d3de83a69533a2997de97ad92989ff78ee

5 years agoDeprecated type() -> scalar_type()
Gao, Xiang [Sat, 30 Mar 2019 17:50:48 +0000 (10:50 -0700)]
Deprecated type() -> scalar_type()

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/18642

Differential Revision: D14696848

Pulled By: ezyang

fbshipit-source-id: 43d1f86ecee5f6c6c5b70fd7d0e2063c3fc473ab

5 years agoTurn on F401: Unused import warning. (#18598)
Edward Yang [Sat, 30 Mar 2019 15:58:10 +0000 (08:58 -0700)]
Turn on F401: Unused import warning. (#18598)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18598
ghimport-source-id: c74597e5e7437e94a43c163cee0639b20d0d0c6a

Stack from [ghstack](https://github.com/ezyang/ghstack):
* **#18598 Turn on F401: Unused import warning.**

This was requested by someone at Facebook; this lint is turned
on for Facebook by default.  "Sure, why not."

I had to noqa a number of imports in __init__.  Hypothetically
we're supposed to use __all__ in this case, but I was too lazy
to fix it.  Left for future work.

Be careful!  flake8-2 and flake8-3 behave differently with
respect to import resolution for # type: comments.  flake8-3 will
report an import unused; flake8-2 will not.  For now, I just
noqa'd all these sites.

All the changes were done by hand.

Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Differential Revision: D14687478

fbshipit-source-id: 30d532381e914091aadfa0d2a5a89404819663e3

5 years agoUpdate documentation for CTCLoss (#18415)
ryan [Sat, 30 Mar 2019 08:20:55 +0000 (01:20 -0700)]
Update documentation for CTCLoss (#18415)

Summary:
This is meant to resolve #18249, where I pointed out a few things that could improve the CTCLoss docs.

My main goal was to clarify:
- Target sequences are sequences of class indices, excluding the blank index
- Lengths of `target` and `input` are needed for masking unequal length sequences, and do not necessarily = S, which is the length of the longest sequence in the batch.

I thought about Thomas's suggestion to link the distill.pub article, but I'm not sure about it. I think that should be up to y'all to decide.

I have no experience with .rst, so it might not render as expected :)

t-vi ezyang
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18415

Differential Revision: D14691969

Pulled By: soumith

fbshipit-source-id: 381a2d52307174661c58053ae9dfae6e40cbfd46

5 years agoFallback kernels (#18443)
Sebastian Messmer [Sat, 30 Mar 2019 07:03:46 +0000 (00:03 -0700)]
Fallback kernels (#18443)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18443

Allow registering a kernel without a dispatch key. In this case, the kernel becomes a fallback kernel that is called whenever no other kernel matches.
This is also useful for the legacy function based API (since that API doesn't know about dispatch keys) or any other custom ops that don't care about dispatch
and just want one kernel to be called no matter the dispatch key.

Reviewed By: dzhulgakov

Differential Revision: D14603258

fbshipit-source-id: 242dc8871dad2989ca25079854d0cc97429e7199

5 years agoIntroduce lambda-based kernel API (#18541)
Sebastian Messmer [Sat, 30 Mar 2019 07:03:46 +0000 (00:03 -0700)]
Introduce lambda-based kernel API (#18541)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18541

Allow registering lambdas as c10 kernels.

Reviewed By: dzhulgakov

Differential Revision: D14653005

fbshipit-source-id: f867cc776b1339e83b7a2e1935f5cf924cfba44a

5 years agoReport better errors when kernel or dispatch key are missing (#18302)
Sebastian Messmer [Sat, 30 Mar 2019 07:03:46 +0000 (00:03 -0700)]
Report better errors when kernel or dispatch key are missing (#18302)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18302

These might be use cases we want to support in the future, but they don't work yet.
Let's at least report an error instead of doing segfaults or worse.

Reviewed By: dzhulgakov

Differential Revision: D14572346

fbshipit-source-id: 49262ce131493bc887defe2978d8b22f202cd8cc

5 years agoMove stuff to cpp files (#18301)
Sebastian Messmer [Sat, 30 Mar 2019 07:03:44 +0000 (00:03 -0700)]
Move stuff to cpp files (#18301)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18301

Move code out of headers and templates into source files and non-templates.

Reviewed By: dzhulgakov

Differential Revision: D14572347

fbshipit-source-id: 9fd5d62d54000a95e93076cd73f591ba2c5c2653

5 years agoCheck kernel against function schema in c10 op registration (#18256)
Sebastian Messmer [Sat, 30 Mar 2019 07:03:44 +0000 (00:03 -0700)]
Check kernel against function schema in c10 op registration (#18256)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18256

This diff infers the function schema from the kernel function/functor and checks that it matches the specified function schema.

This diff does not allow (yet) to omit specifying the function schema in the registration API. That will come in a future diff.

Reviewed By: dzhulgakov

Differential Revision: D14552738

fbshipit-source-id: 00202b489ede19f26ae686c97416b38c72c11532

5 years agoAdd functor- and function-based kernel registration API (#18162)
Sebastian Messmer [Sat, 30 Mar 2019 07:03:44 +0000 (00:03 -0700)]
Add functor- and function-based kernel registration API (#18162)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18162

- Adds the API to register a functor- and function-based kernel.
- Change the experimental c10 ops to use this new API instead of the old one
- Deletes the old APIs in KernelRegistration.h and OpSchemaRegistration.h

Reviewed By: dzhulgakov

Differential Revision: D14514239

fbshipit-source-id: 35b2f6e8f62964e54886450a6a5fac812ed20f26

5 years agoNew operator registration MVP (#18161)
Sebastian Messmer [Sat, 30 Mar 2019 07:03:43 +0000 (00:03 -0700)]
New operator registration MVP (#18161)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18161

This introduces version 0 for the new operator registration.

For now, it only works with kernels that are defined as stack-based functions.
This is actually not the intended public API for defining kernels, but it's the basis which is going to be used to define the public APIs (see diffs on top for them),
and it's also the API used for exposing caffe2 operators.

This diff also switches the mechanism for exposing caffe2 operators to the new mechanism.

Reviewed By: dzhulgakov

Differential Revision: D14514231

fbshipit-source-id: 454ab7b5b46a10203aa27b175400d23f818dd1df

5 years agoFix trt installation in CI (#18609)
Junjie Bai [Sat, 30 Mar 2019 05:42:18 +0000 (22:42 -0700)]
Fix trt installation in CI (#18609)

Summary:
caffe2_py2_cuda9_0_cudnn7_ubuntu16_04_build is failing
```
...
Mar 29 04:44:46 Need to get 174 MB of archives.
Mar 29 04:44:46 After this operation, 576 MB of additional disk space will be used.
Mar 29 04:44:46 Do you want to continue? [Y/n] Abort.
Exited with code 1
...
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18609

Differential Revision: D14694990

Pulled By: bddppq

fbshipit-source-id: 260446a8650f660a2baf123a3f17efdf0a8d6c64

5 years agoAttribute serialization improvements (#18188)
David Riazati [Sat, 30 Mar 2019 02:06:06 +0000 (19:06 -0700)]
Attribute serialization improvements (#18188)

Summary:
* adds attributes to `ScriptModule.__getattr__` so they can be accessed in Python after re-importing
* full support for all the possible values for an `int64_t`
    * this necessitated a bunch more `pushWhatever` functions, so re-introduced a templated version to cut down on duplicate code
* tests to validate references / value sharing works
* adds `torch.jit.Unpickler` which people can use to de-serialize the pickle files into Python / have a quick reference on how to do this without PyTorch
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18188

Differential Revision: D14527490

Pulled By: driazati

fbshipit-source-id: efd15579cc04aa2e28c4b2c9490d82d849dee559

5 years agosupport pre-convert filter format for mkldnn training mode and change 'OptimizeForIde...
Cheng,Penghui [Sat, 30 Mar 2019 01:51:50 +0000 (18:51 -0700)]
support pre-convert filter format for mkldnn training mode and change 'OptimizeForIdeep' to 'OptimizeForMkldnn' (#15171)

Summary:
For MKL-DNN,the filter data will be reorderd to primitive format, it takes a lot of time.
So the patch provide a method to convert filter format before training.
And "OptimizeForIdeep" will be changed to "OptimizeForMkldnn" in this patch.
 This patch depends on https://github.com/pytorch/pytorch/pull/12866
Pull Request resolved: https://github.com/pytorch/pytorch/pull/15171

Differential Revision: D14590741

Pulled By: yinghai

fbshipit-source-id: 07971c9977edac3c8eec08ca2c39cda639683492

5 years agoTensor construction codemod(raw_mutable_data) (#16373)
Jerry Zhang [Sat, 30 Mar 2019 01:26:07 +0000 (18:26 -0700)]
Tensor construction codemod(raw_mutable_data) (#16373)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/16373

motivation: https://github.com/pytorch/pytorch/pull/12407
This is a manual diff.
most of the fixes should be:

```
auto* Y = Output(0);
Y->Resize(dims);
Y->raw_mutable_data(dtype);
```
-->
```
auto* Y = Output(0, dims, at::dtype(dtype));
```
But there might be other cases.

Reviewed By: dzhulgakov

Differential Revision: D13725460

fbshipit-source-id: 649a4b0e42f62cda1a60171dd9fa3e440dc9dca1