[ci] Release only change: bump macos worker instance type (pytorch#82113

)

* [ci] Release only change: bump macos worker instance type

* Applying bump for nightly

* Add macos-12-xl to actionlint

Fix deserialization of TransformerEncoderLayer (pytorch#81832) (pytor…

…ch#81832) (pytorch#82094)

Summary:
When `activation` is a module, it is not saved directly in the state dictionary but instead in `_modules`. When deserialized, the old version of this code would think that activation was missing and set it to RELU. This version first reconstructions the module and then sees if activation is neither a module nor a function before setting it to RELU.

Pull Request resolved: pytorch#81832
Approved by: https://github.com/kit1980, https://github.com/zrphercule

Test Plan:
contbuild & OSS CI, see https://hud.pytorch.org/commit/pytorch/pytorch/e68583b4d180066b8e4f108e0d23176a2676421c

Test plan from GitHub:
pytorch oss tests

Reviewed By: jeanschmidt, zrphercule

Differential Revision: D38014872

Pulled By: zdevito

fbshipit-source-id: 938079d768f7981ca55eed3c8828b29a92e06f41

Co-authored-by: Zachary DeVito (Meta Employee) <[email protected]>

Update on "Add functorch shards for windows CI"

Test Plan:
- wait for CI

[ghstack-poisoned]

ci: Use SCCACHE_S3_KEY_PREFIX in CI builds

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

[ghstack-poisoned]

[PyTorch][Kineto] add ActivityType.h when USE_KINETO is not set (pyto…

…rch#82028)

Summary:
Pull Request resolved: pytorch#82028

This patch fixes an error "'ActivityType.h' file not found" when `use_kineto()` is false.

## Problem
Even when `use_kineto()` is not set (i.e., `-DUSE_KINETO` is not passed), `ActivityType.h` is required for PyTorch compilation:
https://github.com/pytorch/pytorch/blob/master/torch/csrc/profiler/kineto_shim.h#L15

## Solution
Add `ActivitiyType.h` dependency even when `use_kineto() == False`.

Test Plan: PyTorch internal and external CI tests.

Differential Revision: D38090153

fbshipit-source-id: c9465f7523e4b10ae9bf0fe3c4bc6455f41a6605

Merge branch 'master' into rocm_fused_conv

Update on "Add functorch shard for mac x86 tests, linux cu102 tests"

Test Plan:
- wait for CI

[ghstack-poisoned]

[MPS]: Perf fixes.

Fixes pytorch#81610
* Use fillBuffer() for zero_mps()
Fix minor bug in add_sub_template() with value=0.0
Change default value of use_scalar_value to false in getTensorsStringKey()

* Fallback to fill_scalar_mps() if buffer isn't contiguous.

* Fix high memory consumption in view ops

* Change commitAndWait to Commit in View Ops

Update on "[Profiler] Improve tree testing"

In the course of trying to get pytorch#80797 out the door, I've wound up making a number of minor tweaks to `test_profiler_tree` which collectively have made it much easier to debug. It's also revealed what seems to be a correctness issue with how profiler assigns lineage on certain platforms. So I've decided to pull those testing improvements into a standalone PR.

Differential Revision: [D38038122](https://our.internmc.facebook.com/intern/diff/D38038122/)

[ghstack-poisoned]

Update on "Dispatch the auxiliary frobenius_norm and nuclear_norm to …

…better implementations and deprecate them"

These functions will be legacy functions. We deprecate them, but we also
take this chance to dispatch to a more efficient and consistent implementation.
Doing so should help writing a conversion rule for these to be able to
remove them once and for all

[ghstack-poisoned]