This follows the structure of linalg.svd.

Pull Request resolved: pytorch#79072

Approved by: https://github.com/IvanYashchuk, https://github.com/albanD

Will be skipped when imported internally, for more details see https://www.internalfb.com/diff/D37114156?src_version_fbid=3331368873807344

Pull Request resolved: pytorch#79554
Approved by: https://github.com/albanD

See title

Addresses pytorch#79540

Error it's causing:
```
2022-06-14T16:29:53.6335274Z Results (1120.92s):
2022-06-14T16:29:53.6335495Z      393 passed
2022-06-14T16:29:53.6335710Z        1 failed
2022-06-14T16:29:53.6336041Z          - test/onnx/test_models.py:155 TestModels_new_jit_API.test_inception
2022-06-14T16:29:53.6336326Z       60 skipped
2022-06-14T16:29:54.4670969Z ##[error]Process completed with exit code 1.
2022-06-14T16:29:54.4730658Z Prepare all required actions
2022-06-14T16:29:54.4730993Z Getting action download info
<probably uninteresting folded group, click to show>
```
Pull Request resolved: pytorch#79556
Approved by: https://github.com/janeyx99, https://github.com/seemethere

Signed-off-by: Edward Z. Yang <ezyangfb.com>

Pull Request resolved: pytorch#79493

Approved by: https://github.com/albanD

Signed-off-by: Edward Z. Yang <ezyangfb.com>

Pull Request resolved: pytorch#79494

Approved by: https://github.com/albanD

…rch#79541)

Allows use of .jenkins/pytorch/build.sh without assuming OUR_GITHUB_JOB_ID is set.  This is a regression caused by pytorch#79366.
Pull Request resolved: pytorch#79541
Approved by: https://github.com/suo, https://github.com/seemethere

…orch#76601)

Fixes pytorch#68299.  Fixes pytorch#70875.

Test is flaky on ROCm because the HIP runtime occasionally copies asynchronously too quickly for the current sleep value of 50ms.  This is not a bug.  Increasing the sleep value to 1s to avoid flakiness.
Pull Request resolved: pytorch#76601
Approved by: https://github.com/pruthvistony, https://github.com/malfet

Enabling:
test_sampled_addmm_errors_cuda_complex128
test_sampled_addmm_errors_cuda_complex64
test_sampled_addmm_errors_cuda_float32
test_sampled_addmm_errors_cuda_float64
test_sparse_add_cuda_complex128
test_sparse_add_cuda_complex64

Pull Request resolved: pytorch#77877
Approved by: https://github.com/pruthvistony, https://github.com/malfet

This process is pretty confusing, so wrote it down.

[skip ci]

Pull Request resolved: pytorch#79504

Approved by: https://github.com/janeyx99

This PR is auto-generated nightly by [this action](https://github.com/pytorch/pytorch/blob/master/.github/workflows/_update-commit-hash.yml).
Update the pinned xla hash.
Pull Request resolved: pytorch#79172
Approved by: https://github.com/zengk95

Pull Request resolved: pytorch#79547

Approved by: https://github.com/ejguan

Summary: implemented threshold op for vulkan

Test Plan: buck run //xplat/caffe2:pt_vulkan_api_test_binAppleMac

Differential Revision: D36681867

Pull Request resolved: pytorch#78654
Approved by: https://github.com/SS-JIA

…#79482)

Summary:
- StaticModule was being created at runtime which was adding overhead to the forked operation
- Move staticModule creation to outside of runtime so that StaticRuntime instance can be created on top of same staticModule that is created once

Differential Revision: D37126923

Pull Request resolved: pytorch#79482
Approved by: https://github.com/tenpercent

This PR adds variance function with correction argument to nvFuser.

Now it's possible to run
```py
import torch
import torch._refs
from torch._prims.executor import make_traced

def foo1(a):
    return torch._refs.var(a, keepdim=False, unbiased=False)

def foo2(a):
    return torch._refs.var(a, keepdim=False, correction=2)

a = torch.randn(3, 3, device='cuda')
make_traced(foo1)(a, executor="nvfuser")
make_traced(foo2)(a, executor="nvfuser")
```

Pull Request resolved: pytorch#79517
Approved by: https://github.com/mruberry, https://github.com/jjsjann123

Summary: The _detect_dynamic_vs_static function was added to take in a
prepared fx graph model that already had ModelReportObservers built into
it and uses the collected information to determine whether input and
output are stationary or non-stationary and provides feedback on whether
to make linear modules static or dynamic based on this information.

This PR will be followed up soon with another PR that will more
rigoursly test the whole end to end performance of this system, which is
primarily how the function in this PR will be tested for functionality,
which is why this one only has 1 test.

Test Plan: python test/quantization/fx/test_model_report_fx.py TestModelReportDetectDynamicStatic

Reviewers:

Subscribers:

Tasks:

Tags:

Pull Request resolved: pytorch#79326

Approved by: https://github.com/HDCharles

# Summary

ShipIt jobs triggered by co-development workflows are failing to merge PRs due to stale checks.  This diff skips the stale check when merge is triggered by `facebook-github-bot`.

Sample merge failure: pytorch#78654 (comment)
Pull Request resolved: pytorch#79572
Approved by: https://github.com/bigfootjon, https://github.com/seemethere, https://github.com/malfet

…8088)

Fixes #ISSUE_NUMBER

Pull Request resolved: pytorch#78088
Approved by: https://github.com/JackCaoG, https://github.com/wconstab

This reverts commit 05664a9.

Reverted pytorch#79494 on behalf of https://github.com/ezyang due to conflicts with earlier diff that needs revert

This reverts commit d332724.

Reverted pytorch#78135 on behalf of https://github.com/ezyang due to broke torchvision tests

This reverts commit 6b015af.

Reverted pytorch#79493 on behalf of https://github.com/ezyang due to this land races with a revert

…ster (pytorch#79559)

Relates to pytorch#76700

**Overview**: Wrote GHA to get the latest commit SHA. Another component of the script is pushing this SHA to the viable/strict branch, which I will test on pytorch/pytorch-canary.

Todo in the next PR: add comment explaining cron, replace package installation statements with txt file

**Test Plan:** Monitor github actions results to see if the SHA printed is correct by running GHA on pytorch/pytorch-canary. The successful test workflow is [here](https://github.com/pytorch/pytorch-canary/runs/6888486129?check_suite_focus=true).
Pull Request resolved: pytorch#79559
Approved by: https://github.com/janeyx99

…#79568)

Fixes pytorch#79531

Pull Request resolved: pytorch#79568
Approved by: https://github.com/jbschlosser

Pull Request resolved: pytorch#79545

Approved by: https://github.com/ngimel, https://github.com/albanD

…behavior & warn

Pull Request resolved: pytorch#79549

Approved by: https://github.com/ngimel, https://github.com/albanD

Includes pytorch@30fb2c4 and pytorch@95b15c2

Pull Request resolved: pytorch#79593
Approved by: https://github.com/janeyx99

Remove as many references as can be easily done of unittest in favor of our custom infra.

Left a todo where I could not easily replace unittest.main with run_tests()
Pull Request resolved: pytorch#79546
Approved by: https://github.com/seemethere

Relates to pytorch#76700

**Overview**: One edge case not accounted for in the original logic of `isGreen` was for commits with no workflow checks. Similarly, if any of the required checks are not present (ex: if all of the pull checks are skipped), the workflow should not be promoteble. A commit should only be promoteable if there is it least one workflow check from each required group present (i.e. none of them are skipped)

**Test Plan:** Verify that commits on the HUD with no workflow checks are not considered promote-able. Added a test case with no workflows in `test_print_latest_commits.py`
Pull Request resolved: pytorch#79565
Approved by: https://github.com/seemethere

see title
```
2022-06-15T15:20:12.5183743Z The PR is introducing backward incompatible changes to the operator library. Please contact PyTorch team to confirm whether this change is wanted or not.
2022-06-15T15:20:12.5183765Z
2022-06-15T15:20:12.5183880Z Broken ops: [
2022-06-15T15:20:12.5184275Z 	aten::empty.SymInt(SymInt[] size, *, int? dtype=None, int? layout=None, Device? device=None, bool? pin_memory=None, int? memory_format=None) -> Tensor
2022-06-15T15:20:12.5184342Z ]
2022-06-15T15:20:12.6303306Z + cleanup
2022-06-15T15:20:12.6303395Z + retcode=1
2022-06-15T15:20:12.6303463Z + set +x
2022-06-15T15:20:12.6345211Z ##[error]Process completed with exit code 1.
2022-06-15T15:20:12.6377724Z Prepare all required actions
2022-06-15T15:20:12.6377844Z Getting action download info

```
Pull Request resolved: pytorch#79612
Approved by: https://github.com/malfet

)"

This reverts commit 4ebb326.

Reverted pytorch#79034 on behalf of https://github.com/facebook-github-bot due to Diff reverted internally

Moves the deply job prefix to a suffix to make it more consistent with
our other job names

Signed-off-by: Eli Uriegas <eliuriegasfb.com>

Pull Request resolved: pytorch#79436

Approved by: https://github.com/janeyx99

…#79298)

Adds a `same_shape` util and updates maybe_broadcast to use it; previously maybe_broadcast was always broadcasting because its equality check was always failing.
Pull Request resolved: pytorch#79298
Approved by: https://github.com/ezyang

…t from the middle (pytorch#75763)

Summary: When 3 tensor lists are packed into variadic arg list, it is convenient to view the original lists as (start, end) pairs in the varargs list

Differential Revision: D35622049

Pull Request resolved: pytorch#75763
Approved by: https://github.com/mikeiovine

What was happening is that when we have multiple learning rate schedulers, the order in which they are being initialized is not being taken into account. This is a problem if they were being initialized in sequential order (as one might intuitively do).

Each scheduler calls `step()` on initialization and sets the `lr` in its optimizer's `params_groups`. However, this means that step 0 will be using the `lr` that was set by the very last scheduler (in the case of initializing schedulers sequentially) instead of the first scheduler.

The fix in this PR, addresses the above bug by performing a call to the appropriate scheduler on initialization after decrementing the `last_epoch` values in order to keep them the same post-step. This will ensure that the correct scheduler is the one setting the `lr` values for the optimizer's `param_groups`
Pull Request resolved: pytorch#72856
Approved by: https://github.com/jbschlosser

Update release.md with rc validation steps

Pull Request resolved: pytorch#79889
Approved by: https://github.com/seemethere

Summary: The use of `c10::nullopt` instead of `c10::optional<Tensor>()` caused a null pointer dereference on Android.

Test Plan:
On Mac:
```
buck run //xplat/caffe2:pt_vulkan_api_test_binAppleMac
```

On Android:
```
buck build -c ndk.custom_libcxx=false -c pt.enable_qpl=0 //xplat/caffe2:pt_vulkan_api_test_binAndroid\#android-arm64 --show-output
adb push buck-out/gen/xplat/caffe2/pt_vulkan_api_test_binAndroid\#android-arm64 /data/local/tmp/vulkan_api_test
adb shell "/data/local/tmp/vulkan_api_test"
```

Reviewed By: SS-JIA

Differential Revision: D37189192

Pull Request resolved: pytorch#79701
Approved by: https://github.com/SS-JIA

Last week, we had to disable it due to infra issues. We are now re enabling since the infra issues are past.

Test Plan:
Reran this job and it looks OK.
Pull Request resolved: pytorch#79949
Approved by: https://github.com/seemethere

Pull Request resolved: pytorch#79874

Approved by: https://github.com/robieta

We have hard limitations on the number of linux.16xlarge.nvidia.gpu
machines we can spin up. Considering that the TTS for this specific job
has increased 2x over the past 7 days.

Signed-off-by: Eli Uriegas <eliuriegasfb.com>

Pull Request resolved: pytorch#79894

Approved by: https://github.com/malfet, https://github.com/janeyx99

…79942)

Summary:
Memory planner destruction was hitting [this assertion](https://www.internalfb.com/code/fbsource/[f8baf8a0bab462c860d2eb7491a4e3fb40d2907a]/fbcode/caffe2/c10/util/intrusive_ptr.h?lines=117) in debug mode for a few models.

Here's what was going on:

1) The set of unmanaged `IValue`s acquires one or more owning refs of a managed `StorageImpl`
2) Then, one or more tensors in that storage group have their `StorageImpl` swapped out during execution
3) During `deallocateManagedTensors`, we swap the correct `StorageImpl` back in, [calling `unsafe_adapt_non_heap_allocated` again and resetting the refcount](https://www.internalfb.com/code/fbsource/[f8baf8a0bab462c860d2eb7491a4e3fb40d2907a]/fbcode/caffe2/torch/csrc/jit/runtime/static/memory_planner.cpp?lines=446-452)
4) The unmanaged `IValues` are deallocated, decrementing the refcount into the danger zone.

So, we just have to make sure that unmanaged `IValue`s are destructed before we deallocate the managed tensors.

Test Plan: CI

Differential Revision: D37303728

Pull Request resolved: pytorch#79942
Approved by: https://github.com/tenpercent

…#79905)

We now use Rockset to store test case stats and we can glean flakiness from that data, so this complicated logic isn't necessary anymore!

This will also prevent spurious red on trunk that occurs with this code e.g., https://github.com/pytorch/pytorch/runs/6968080598?check_suite_focus=true

## test plan
print_test_stats work on the PR: https://github.com/pytorch/pytorch/runs/6974961237?check_suite_focus=true though it seems several jobs ran into HTTP 500 and Memory errors which I don't think is related to my code + more a problem with the server https://github.com/pytorch/pytorch/runs/6974938954?check_suite_focus=true

Pull Request resolved: pytorch#79905
Approved by: https://github.com/zengk95, https://github.com/seemethere

…torch#79945)

###  Motivation

In order to match the internal platform010 builds, we are creating a new config to run on PRs that uses compiler and package versions >= used in platform010. Here are the versions used in the new build

- Ubuntu 22.04 (Jammy-Jellyfish)
- Clang-12
- Python 3.8
- CUDA 11.6

### Summary of Changes

- As `nvidia/docker` images only support Cuda 11.7 with Ubuntu 22.04, we are starting with base Ubuntu 22.04 docker images, and then installing Cuda 11.6

- Fetching `install_cuda.sh` from `pytorch/builder` repo in order to install Cuda using `wget`

- `libssl-dev` has been upgraded to libssl3 in Ubuntu 22.04. Instead, we are using `include` and `lib` folders downloaded with `Openssl1.1.1`

- `Clang-12` requires `libomp-12-dev` to work with `OpenMP` which is added to the `install_base.sh` file.

- Minor fixes to handle compilation errors generated when using `clang-12`.
      - In `pow_test.cpp` adding a `static_cast` to input of `sqrt` method
      - In `vec512_qint.h`, explicitly defining `copy-assignment` operator as its implicit definition is deprecated due to
         user-declared `copy-constructor` in C++11
Pull Request resolved: pytorch#79945
Approved by: https://github.com/seemethere, https://github.com/kit1980

Ref: pytorch#69991
Pull Request resolved: pytorch#78515
Approved by: https://github.com/zou3519, https://github.com/Lezcano

For ignored modules' parameters, we should also clean their parameter names since they will have the FSDP-specific prefixes.

This change only affects the prefixed parameter name keys in `full_optim_state_dict()` (i.e. optim state dict saving). Not having this change does not actually violate the correctness of the optim state dict save-load flow because it only requires that the keys are unique and internally consistent.

Either way, this PR explicitly adds the specification now that the parameter keys in the optim state dict should match the keys of full model state dict.
Pull Request resolved: pytorch#79955
Approved by: https://github.com/rohan-varma

Otherwise it errors on ONNX clang10 build: https://github.com/pytorch/pytorch/runs/6976721937?check_suite_focus=true
Pull Request resolved: pytorch#79964
Approved by: https://github.com/malfet

…utputs

Without profiled outputs, autodiff can't tell whether or not the outputs of a DifferentiableGraph should requires_grad. Autodiff would default to requires_grad=True if there was no profiled information, causing autodiff to mark tensors as requires_grad when they shouldn't have. This adds requires_grad info onto the type of the output, if it can be found in later uses of the output.

Adds a test for correct autodiff requires_grad behavior and also a test to make sure the output type is correctly annotated in create_autodiff_subgraphs.

Pull Request resolved: pytorch#79498

Approved by: https://github.com/eellison

Fixes #ISSUE_NUMBER

Pull Request resolved: pytorch#79947
Approved by: https://github.com/ngimel

Fixes pytorch#79871

Make `module.cpp` tests respect change that was made in pytorch#78436 (no int types in autograd).

Note that there still a gap in Cmake test -- it's unclear why it didn't fail CI before.

As far as I can tell it should be executed, because it's included here https://github.com/pytorch/pytorch/blob/79507d2a9d06d4a3fb50eb21b30e08cc044776ce/test/cpp/api/CMakeLists.txt#L17:L17

Pull Request resolved: pytorch#79926
Approved by: https://github.com/soulitzer

…torch#74704)"

This reverts commit 93b0fec.

Reverted pytorch#74704 on behalf of https://github.com/malfet due to broke torchvision

per title
Pull Request resolved: pytorch#79973
Approved by: https://github.com/mruberry

Currently, `bitonicSortKVInPlace` is written to sort one array per
block of threads. If that dimension happens to be very small
(<128 elements), this results in low thread occupancy.

Instead, this changes `bitonicSortKVInPlace` to operate with a 2d
block. Sorting happens along the x dimension, and the y dimension
is a fixed size batch.

Pull Request resolved: pytorch#79627

Approved by: https://github.com/ngimel