[PyTorch] faster empty_tensor_restride via smarter refresh_contiguous

[swolchok]

Stack from ghstack:

• [PyTorch] Use .sizes() isntead of .size() in cat_serial_kernel_impl #49371 [PyTorch] Use .sizes() isntead of .size() in cat_serial_kernel_impl
• [PyTorch] Use .sizes() instead of .size() in _cat_out_cpu #49368 [PyTorch] Use .sizes() instead of .size() in _cat_out_cpu
• [PyTorch] Avoid extra Tensor refcounting in _cat_out_cpu #49364 [PyTorch] Avoid extra Tensor refcounting in _cat_out_cpu
• [PyTorch] Merge CoinflipTLS into RecordFunctionTLS #49359 [PyTorch] Merge CoinflipTLS into RecordFunctionTLS
• [PyTorch] Avoid move-constructing a List in listConstruct #49355 [PyTorch] Avoid move-constructing a List in listConstruct
• [PyTorch] faster empty_tensor_restride via smarter refresh_contiguous #49349 [PyTorch] faster empty_tensor_restride via smarter refresh_contiguous
• [PyTorch] Make tls_local_dispatch_key_set inlineable #49264 [PyTorch] Make tls_local_dispatch_key_set inlineable
• [PyTorch] Make TORCH_CHECK less likely to interfere with inlining #49263 [PyTorch] Make TORCH_CHECK less likely to interfere with inlining
• [PyTorch] native_cpp_binding for size() and stride() #49262 [PyTorch] native_cpp_binding for size() and stride()

empty_tensor_restride knows what memory format it just
restrided (restrode?) to. Don't throw that information away.

Differential Revision: D25539840

NOTE FOR REVIEWERS: This PR has internal Facebook specific changes or comments, please review them on Phabricator!

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit ef5c949 (more details on the Dr. CI page):

---

Commit ef5c949 was recently pushed. Waiting for builds...

---

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

Follow this link to opt-out of these comments for your Pull Requests.

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

This comment has been revised 1 time.

[ngimel]

is_coniguous_ and is_channels_last_contiguous_ are not mutually exclusive, e.g. contiguous tensor of size (N,C,1,1) has both as true.

[swolchok]

is there anything useful about this diff, then? I don't understand all the subtleties here.

[ngimel]

There are useful things, contiguous computations are repeated twice now, and they shouldn't be. But you are right, it's more annoying. I should still do it.

[swolchok]

abandoning, not as obvious as I thought