[PyTorch] Devirtualize TensorImpl::dim() with macro

[swolchok]

Stack from ghstack:

• [PyTorch] Devirtualize TensorImpl::sizes() with macro #50176 [PyTorch] Devirtualize TensorImpl::sizes() with macro
• [PyTorch] Devirtualize TensorImpl::dim() with macro #49770 [PyTorch] Devirtualize TensorImpl::dim() with macro
• [PyTorch] validate that SparseTensorImpl::dim needn't be overridden #49767 [PyTorch] validate that SparseTensorImpl::dim needn't be overridden
• [PyTorch] Devirtualize TensorImpl::numel() with macro #49766 [PyTorch] Devirtualize TensorImpl::numel() with macro

Seems like the performance cost of making this commonly-called method virtual isn't worth having use of undefined tensors crash a bit earlier (they'll still fail to dispatch).

Differential Revision: D25687465

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 d7e578f (more details on the Dr. CI page):

---

• 2/2 failures possibly* introduced in this PR
  • 1/2 non-CircleCI failure(s)

---

🕵️ 1 new failure recognized by patterns

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

pytorch_libtorch_linux_xenial_cuda11_1_cudnn8_py3_gcc7_build (1/1)

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

Jan 07 18:06:41 sccache: error: couldn't connect to server

Jan 07 18:06:41 +++ eval 'extract_trap_cmd '
Jan 07 18:06:41 ++++ extract_trap_cmd
Jan 07 18:06:41 ++++ printf '%s\n' ''
Jan 07 18:06:41 +++ printf '%s\n' cleanup
Jan 07 18:06:41 ++ trap -- '
Jan 07 18:06:41 cleanup' EXIT
Jan 07 18:06:41 ++ [[ pytorch-libtorch-linux-xenial-cuda11.1-cudnn8-py3-gcc7-build != *pytorch-win-* ]]
Jan 07 18:06:41 ++ which sccache
Jan 07 18:06:41 ++ sccache --stop-server
Jan 07 18:06:41 Stopping sccache server...
Jan 07 18:06:41 sccache: error: couldn't connect to server
Jan 07 18:06:41 sccache: caused by: Connection refused (os error 111)
Jan 07 18:06:41 ++ true
Jan 07 18:06:41 ++ rm /var/lib/jenkins/sccache_error.log
Jan 07 18:06:41 rm: cannot remove '/var/lib/jenkins/sccache_error.log': No such file or directory
Jan 07 18:06:41 ++ true
Jan 07 18:06:41 ++ [[ pytorch-libtorch-linux-xenial-cuda11.1-cudnn8-py3-gcc7-build == *rocm* ]]
Jan 07 18:06:41 ++ SCCACHE_ERROR_LOG=/var/lib/jenkins/sccache_error.log
Jan 07 18:06:41 ++ SCCACHE_IDLE_TIMEOUT=1200
Jan 07 18:06:41 ++ RUST_LOG=sccache::server=error
Jan 07 18:06:41 ++ sccache --start-server

---

1 job timed out:

• pytorch_libtorch_linux_xenial_cuda11_1_cudnn8_py3_gcc7_build

---

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 33 times.

[ezyang]

OK. This change is a little hard to understand from the perspective of an OSS only user, since there is no point in the codebase where we ever actually set the macro. A Note explaining the situation would go a long way to explaining to people what's going on. (It's also worth pointing out XLA is the blocker for unconditionally removing the macro entirely)

[swolchok]

A Note explaining the situation

Added to the PR at the bottom of the stack that adds TENSORIMPL_MAYBE_VIRTUAL.

[facebook-github-bot]

This pull request has been merged in 4de6b27.

[facebook-github-bot]

This pull request has been reverted by c215ffb.