Move CompositeCompliance tests to their own TestCase

[zou3519]

Stack from ghstack:

• Composite Compliance testing for backward formulas #74646
• Test case where some inputs are Tensor Subclasses in CompositeCompiance #74645
• -> Move CompositeCompliance tests to their own TestCase #74644

This is in preparation for me adding additional tests for:

1. composite compliance of autograd formulas
2. composite compliance of forward-mode AD formulas

This PR also changes these tests to run on both CPU and CUDA. Previously
they were just run on CPU, but it turns out there's a lot of branching
on the device in composite operations in PyTorch today :/

Test Plan:

• wait for tests

Differential Revision: D35186861

[facebook-github-bot]

🔗 Helpful links

• 🧪  See artifacts and rendered test results at hud.pytorch.org/pr/74644
• ↩️  [fb-only] Re-run with SSH instructions
• Need help or want to give feedback on the CI? Visit our office hours

💊 CI failures summary and remediations

As of commit 12b6b3b (more details on the Dr. CI page):

---

• 1/1 failures introduced in this PR

---

🕵️ 1 new failure recognized by patterns

The following CI failures do not appear to be due to upstream breakages:

pull / linux-bionic-rocm4.5-py3.7 / test (default, 2, 2, linux.rocm.gpu) (1/1)

Step: "Test" (full log | diagnosis details | 🔁 rerun)

2022-03-28T16:38:03.4735976Z FAIL [0.013s]: test_caching_pinned_memory (__main__.TestCuda)

2022-03-28T16:38:03.4652982Z   test_scatter_cpu_dim (__main__.TestCudaComm) ... skip: only one GPU detected (0.001s)
2022-03-28T16:38:03.4659469Z   test_scatter_cpu_neg_dim (__main__.TestCudaComm) ... skip: only one GPU detected (0.001s)
2022-03-28T16:38:03.4667141Z   test_scatter_cpu_sizes (__main__.TestCudaComm) ... skip: only one GPU detected (0.001s)
2022-03-28T16:38:03.4674968Z   test_scatter_gpu (__main__.TestCudaComm) ... skip: only one GPU detected (0.001s)
2022-03-28T16:38:03.4682177Z   test_scatter_gpu_dim (__main__.TestCudaComm) ... skip: only one GPU detected (0.001s)
2022-03-28T16:38:03.4689863Z   test_scatter_gpu_neg_dim (__main__.TestCudaComm) ... skip: only one GPU detected (0.001s)
2022-03-28T16:38:03.4696879Z   test_scatter_gpu_sizes (__main__.TestCudaComm) ... skip: only one GPU detected (0.001s)
2022-03-28T16:38:03.4730379Z   test_scatter_namedtuple (__main__.TestCudaComm) ... skip: Test needs multiple GPUs (0.003s)
2022-03-28T16:38:03.4731798Z 
2022-03-28T16:38:03.4732109Z ======================================================================
2022-03-28T16:38:03.4735976Z FAIL [0.013s]: test_caching_pinned_memory (__main__.TestCuda)
2022-03-28T16:38:03.4737806Z ----------------------------------------------------------------------
2022-03-28T16:38:03.4738748Z Traceback (most recent call last):
2022-03-28T16:38:03.4740203Z   File "/opt/conda/lib/python3.7/site-packages/torch/testing/_internal/common_utils.py", line 1780, in wrapper
2022-03-28T16:38:03.4741234Z     method(*args, **kwargs)
2022-03-28T16:38:03.4742382Z   File "test_cuda.py", line 1370, in test_caching_pinned_memory
2022-03-28T16:38:03.4743912Z     self.assertNotEqual(t.data_ptr(), ptr, msg='allocation re-used too soon')
2022-03-28T16:38:03.4745640Z   File "/opt/conda/lib/python3.7/site-packages/torch/testing/_internal/common_utils.py", line 2186, in assertNotEqual
2022-03-28T16:38:03.4746807Z     self.assertEqual(x, y, msg, atol=atol, rtol=rtol, **kwargs)
2022-03-28T16:38:03.4748091Z AssertionError: AssertionError not raised : allocation re-used too soon
2022-03-28T16:38:03.4748736Z 

---

This comment was automatically generated by Dr. CI (expand for details).

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

[zou3519]

Context for reviewers: I'm beefing up composite compliance testing that was introduced as #65819 and retagging all of you as reviewers.

[albanD]

SGTM then

[zou3519]

@zou3519 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.