Register et_copy device-copy ops for ATen-mode runtimes

[shoumikhin]

Summary:
The device-placement pass inserts et_copy::_h2d_copy.out / _d2h_copy.out at
CPU<->accelerator boundaries (e.g. CUDA-delegated subgraphs). These portable ops
are registered for the portable runtime via functions.yaml, but ATen-mode
consumers hit "Missing operator: et_copy::_h2d_copy.out" at runtime: ATen-mode
codegen only registers ATen native_functions.yaml ops, so generated_lib_aten
never emits these custom ops and no op__device_copy_aten target existed.

This adds an ATen-mode registration path:

• et_copy_ops.yaml custom-ops schema + device_copy_ops_aten_lib (+_static)
  that registers the ops for ATen-mode runtimes, reusing the op__device_copy
_aten kernel.
• op__device_copy moved into CUSTOM_OPS to produce the _aten target; it
  stays registered for the portable runtime via functions.yaml.
• op__device_copy.cpp gains an ATen-mode kernel branch: the planned at::Tensor
  carries no device metadata in ATen mode, so direction comes from op identity
  (_h2d/_d2h) and the copy uses the registered CUDA DeviceAllocator on the
  current device. The portable branch is factored to share device-accessor
  helpers but keeps its existing behavior.

Scope/limitations:

• ATen-mode device copy is CUDA-only and single-GPU (current device): the
  planned device type/index are not recoverable from the at::Tensor, so the
  kernel targets the registered CUDA allocator's current device. Multi-GPU /
  non-CUDA support is a follow-up (see the TODO in op__device_copy.cpp).

fbcode and xplat copies are kept identical.

Differential Revision: D111752990

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20902

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• No jobs running on PR - "maximum number of queued jobs has been exceeded"

❌ 34 Pending, 1 Unclassified Failure

As of commit 7fc0c28 with merge base ae754e9 ():

UNCLASSIFIED FAILURE - DrCI could not classify the following job because the workflow did not run on the merge base. The failure may be pre-existing on trunk or introduced by this PR:

• trunk / test-llama-runner-mac (fp32, coreml) / macos-job (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 128

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[meta-codesync]

@shoumikhin has exported this pull request. If you are a Meta employee, you can view the originating Diff in D111752990.

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

[Copilot]

Pull request overview

This PR fixes ATen-mode runtime failures for et_copy::_h2d_copy.out / et_copy::_d2h_copy.out by adding an ATen-mode registration path (custom-op schema + generated registration lib) and extending op__device_copy with an ATen-mode kernel implementation.

Changes:

• Add et_copy_ops.yaml plus device_copy_ops_aten_lib to register the device-copy ops in ATen-mode runtimes.
• Adjust operator target wiring so op__device_copy_aten is produced and can be linked by the ATen-mode registration library.
• Implement and test an ATen-mode kernel branch for device copies (direction determined by op identity; uses registered CUDA allocator on current device).

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
shim_et/xplat/executorch/kernels/portable/op_registration_util.bzl	Moves op__device_copy into the list that produces an _aten target as well.
kernels/test/targets.bzl	Adds a dedicated ATen-mode device-copy unit test target.
kernels/test/op__device_copy_aten_test.cpp	New ATen-mode tests validating direction selection, resize behavior, and guardrails.
kernels/portable/targets.bzl	Exports et_copy_ops.yaml, adds an operator library for it, and adds device_copy_ops_aten_lib for ATen-mode registration.
kernels/portable/et_copy_ops.yaml	Defines custom-op schemas for et_copy device-copy ops for ATen-mode codegen/registration.
kernels/portable/cpu/op__device_copy.cpp	Adds ATen-mode kernel path and refactors device metadata access helpers.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated 5 comments.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.

[digantdesai]

Do we need any CMake changes?