Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21531
ghimport-source-id: 711867e19cc3948a5e2a6aa8c4f2cd631abb04d2

Reviewed By: zdevito

Differential Revision: D15719260

Pulled By: suo

fbshipit-source-id: e88c5d3e14e6ecc956ce30ab0246ed606f4b0a38

Summary:
namedtensor build + test should run on PRs only if the commit message
includes [namedtensor ci].
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21520

Differential Revision: D15718404

Pulled By: zou3519

fbshipit-source-id: ce8b5df2682e795e64958a9d49e2e3c091599b33

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21484

cast<int32_t*> => cast<int32_t>

Also fixed reserve problem which might cause incorrect pointer.

Reviewed By: yinghai

Differential Revision: D15699866

fbshipit-source-id: 374418476bddd60f5c5306c8c57319ccf28b9990

Differential Revision:
D15717337

Original commit changeset: 57e65a679a8f

fbshipit-source-id: f73794087a23d56d03497b29d9a9e4e7d54deaad

Summary:
This should further reduce noise by only clang-formatting the lines you actually touched in the precommit hook.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/15657

Differential Revision: D15717337

Pulled By: suo

fbshipit-source-id: 57e65a679a8fdee5c3ff28e241c74ced9398eb0c

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21497
ghimport-source-id: bc03f274

Differential Revision: D15715478

Pulled By: gchanan

fbshipit-source-id: 90e1b65249b4b12f936ee8877cc0bc5a972d9ceb

Summary:
lower tuples pass didn't check bounds for tuple index
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21521

Differential Revision: D15716813

Pulled By: eellison

fbshipit-source-id: 8eead98c2c63118e7d24a8c8bf6184b02afb7dcd

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21343

Needed to binarise features

Reviewed By: yinghai

Differential Revision: D15625653

fbshipit-source-id: 52f48259a040dac35a7000bb1eea9feb5c7ef1ab

Reviewed By: yns88

fbshipit-source-id: 5778cdb5173fc16e5d5474fefa2ea89264101184

Summary:
The new implementation of tracing supports more module. So many error-handling code can be removed by placing the old one (LegacyTracedModule).

cc orionr
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21339

Reviewed By: natalialunova

Differential Revision: D15695154

Pulled By: orionr

fbshipit-source-id: af7d35754e9f34bd1a0ad7b72a9ebe276ff8ab98

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21474
ghimport-source-id: b2477765362248a80557d1a20db02a1290bdcde3

Differential Revision: D15699700

Pulled By: fbhuba

fbshipit-source-id: 1aa4309fec0982c8477cfab29ca5f42d2b171f97

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21496
ghimport-source-id: d7197bcc

Differential Revision: D15715479

Pulled By: gchanan

fbshipit-source-id: fa59eb808d26119b33eb97bb90ef70e95e58458d

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/20533

Reviewed By: zrphercule

Differential Revision: D15579713

Pulled By: houseroad

fbshipit-source-id: 91f3ac0cb14ef226f980362b0013b6b92cb8b8da

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21476
ghimport-source-id: adfd08b8

Differential Revision: D15715270

Pulled By: ezyang

fbshipit-source-id: dde02579d9960ac960306d0a024b8e17846ae0ff

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/21488

Differential Revision: D15715264

Pulled By: ezyang

fbshipit-source-id: 86978f294720e0ce6f60b748a71f0604d6cfa00c

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21466
ghimport-source-id: 0a235c8b

Differential Revision: D15698096

Pulled By: ezyang

fbshipit-source-id: 1759c2681071e9c7e83de3de86daf4333c5f8f3a

Summary:
Fixes #12259, needs to make sure tests (see #13766) don't break due to numerical precision issues. Not sure what would need to be adjusted here...
Pull Request resolved: https://github.com/pytorch/pytorch/pull/13774

Differential Revision: D15715021

Pulled By: ezyang

fbshipit-source-id: 20ce2beee1b39ebe9f023c5f2b25be53acccb5f3

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21281
ghimport-source-id: 4b241d54

Differential Revision: D15699063

Pulled By: zou3519

fbshipit-source-id: c0f00c370d266a4ea5211aae943041fd899e960a

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21280
ghimport-source-id: 92184832

Differential Revision: D15698516

Pulled By: zou3519

fbshipit-source-id: 502b9b019d51dd46327e6caf2af69aa520c70cb6

Revert "Revert D15632268: [pytorch][PR] Continuation of Port max_unpool1d, max_unpool2d and max_unpool3d to ATen" (#21427)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21427
ghimport-source-id: 930c2fb2

Differential Revision: D15698423

Pulled By: ezyang

fbshipit-source-id: 891c94c24b6d377cd6dd94d86cc66465b582359f

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21479
ghimport-source-id: 60fa97fb

Differential Revision: D15713609

Pulled By: ezyang

fbshipit-source-id: a3d9c49e2db985f4373508cd44e94d43ae6e24da

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21492

If one async operator failed, async_scheduling net currently only marks all scheduled async operators as finished without cancelling the callbacks.

The new behavior is to cancel the callbacks first, then set event status to finished.

Reviewed By: ilia-cher

Differential Revision: D15702475

fbshipit-source-id: 55a1774d768b2e238bab859b83332f1877a001ca

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21423

- add gradient for boolean mask
- add test for gradient checking

Reviewed By: BIT-silence

Differential Revision: D15640036

fbshipit-source-id: 79f40c6901e805bf1b8e9b01b57903e30b00f654

Reviewed By: yns88

fbshipit-source-id: af5812e3d071e66f9d0272c36bf639eb04bde7e4

Summary:
This saves ~7% DenseNet load time (4.3s -> 4.0s) on my laptop.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21429

Differential Revision: D15681374

fbshipit-source-id: 9925a6154d51f2d592e26cb5ff8bf7ab3ee2519b

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/21444

Differential Revision: D15701786

Pulled By: ezyang

fbshipit-source-id: 8348e08f9b8f3047b30736f9a944786ab84e6b68

Summary:
Turing GPUs (compute capability 7.5) require CUDA10 to work properly.
We've seen some issues for these GPUs using PyTorch binaries with CUDA9 or older:
[Discussion Board #1](https://discuss.pytorch.org/t/cudnn-status-execution-failed-error/38575)
[Discussion Board #2](https://discuss.pytorch.org/t/cublas-runtime-error-on-gpu-running-but-works-on-cpu/46545/6)

Tested on using CUDA9 with an RTX 2080Ti.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21468

Differential Revision: D15696170

Pulled By: ezyang

fbshipit-source-id: ed43f4e4948d3f97ec8e7d7952110cbbfeafef2a

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21470
ghimport-source-id: 69800c1ce1187591b7bcdb8a63973b4fd8d0e326

Differential Revision: D15696930

Pulled By: ezyang

fbshipit-source-id: fafbcba38d9572a23ee9c1d81cdcce3a154ae4c6

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21502

In BenchResult, we keep name, avg_fwd, std_fwd, avg_bwd, and std_bwd. There is no information about the number of each iteration. In this diff, I am adding more info to BenchResult to include the number reported from each iteration.

Reviewed By: wanchaol

Differential Revision: D15706306

fbshipit-source-id: 3f14be4ba91f1f6da473995783bd7af1d067938d

Differential Revision:
D15629687

Original commit changeset: 2f87f18be655

fbshipit-source-id: a142c22be3fdf14a2b3c29b8766b218fb0883927

Summary:
This moves `JitTestCase` to its own file so that we can have other jit
test files (ex. `test_jit_py3.py`)

There aren't any code changes, just a move and cleaning up the imports
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21491

Pulled By: driazati

Differential Revision: D15703060

fbshipit-source-id: 6082e8b482100bb7b0cd9ae69738f1273e626171

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21426

-

Differential Revision: D15679789

fbshipit-source-id: 5fd448e66af159fd79883aa874065424ec9694ad

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21368

-

Differential Revision: D15629687

fbshipit-source-id: 2f87f18be65552f3eb3f4c945d7f19ba4bae0eb8

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21230

tsia; we support empty tensor with this diff for reshape operator

Reviewed By: jerryzh168

Differential Revision: D15583356

fbshipit-source-id: 6d44c04e95ca3546509bfb12102e29c878f9a7c7

Summary:
Modify MKLDNN pooling operation to support ceil mode by adjusting the right/bottom padding accordingly. This is done similarly as in Caffe (see discussion https://github.com/pytorch/pytorch/pull/19205#discussion_r276903751).

To make this possible, I split the padding to left and right (top / bottom). This naming is confusing but actually follows mkldnn's own naming for pooling::compute(). We increase the r paddings so that it matches the ceiling mode expected output size.

Strengthened the test case.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21310

Reviewed By: bddppq

Differential Revision: D15611664

Pulled By: akyrola

fbshipit-source-id: 46b40015dafef69a8fd5e7b2c261d8dbf448cd20

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21393

Result of splitting the base diff. We moved a header from src/* to include/fbgemm/*

Reviewed By: jianyuh

Differential Revision: D15635188

fbshipit-source-id: ad7d0ddba964ff1cb8b2e33f5f98e457a4d2eac9

Summary:
https://github.com/pytorch/pytorch/issues/18111
Pull Request resolved: https://github.com/pytorch/pytorch/pull/20458

Differential Revision: D15699732

Pulled By: nairbv

fbshipit-source-id: f7a5424c1f1d3b0e4eba0d503d75ae8a18ef7ff4

Summary:
changed `UpsampleBilinearKernel` s.t. the throughput increased 40~50%.

I tested locally with my local test code -- **not pytorch's provided test code** -- because I am having a build problem ( which I made an issue about [here](https://github.com/pytorch/pytorch/issues/19184)). I tested with various tensor sizes and across all the sizes, it should a significant increase in throughput.

1. added `__restrict__`
2. instead of launch as many threads as there are output elements, I launched only `output_height * output_width` may threads and had each thread iterate through the channel and batch dimension.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/19306

Differential Revision: D15701840

Pulled By: ezyang

fbshipit-source-id: 53c54d4f4e4a28b58ecc7d7ae6b864cbfc760e27

Summary:
fix #18254 for numerically instability of `SigmoidTransform`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/19802

Differential Revision: D15701837

Pulled By: ezyang

fbshipit-source-id: fe6c755c523487c8bbdcc3bfb8455801617c70a4

Summary:
Currently, when the input of MVN is precision matrix, we take inverse to convert the result to covariance matrix. This, however, will easily make the covariance matrix not positive definite, hence will trigger a cholesky error.

For example,
```
import torch
torch.manual_seed(0)
x = torch.randn(10)
P = torch.exp(-(x - x.unsqueeze(-1)) ** 2)
torch.distributions.MultivariateNormal(loc=torch.ones(10), precision_matrix=P)
```
will trigger `RuntimeError: cholesky_cpu: U(8,8) is zero, singular U.`

This PR uses some math tricks ([ref](https://nbviewer.jupyter.org/gist/fehiepsi/5ef8e09e61604f10607380467eb82006#Precision-to-scale_tril)) to only take inverse of a triangular matrix, hence increase the stability.

cc fritzo, neerajprad , SsnL
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21366

Differential Revision: D15696972

Pulled By: ezyang

fbshipit-source-id: cec13f7dfdbd06dee94b8bed8ff0b3e720c7a188