Add stable::Tensor.device()

[mikaylagawarecki]

Stack from ghstack (oldest at bottom):

• Redo add parallel_for to torch/csrc/stable #166695
• -> Add stable::Tensor.device() #166694
• Add torch::stable::Device #166579
• Add scaffolding for StableIValue FC/BC (no PoC) #164332
• Add scaffolding for aoti_torch_call_dispatcher BC with native ops #163683
• Move static from_ivalue/to_ivalue to new shim_common.cpp #166373
• Add TORCH_TARGET_VERSION for stable ABI #164356

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/166694

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki

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:

• ROCm failures during provisioning step due to network issues

✅ You can merge normally! (4 Unrelated Failures)

As of commit 7b6869c with merge base 4295a9a ():

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

• pull / linux-jammy-py3.10-clang12 / test (default, 3, 5, linux.4xlarge) (gh) (disabled by #156921 but the issue was closed recently and a rebase is needed to make it pass)
  test/dynamo/test_structured_trace.py::StructuredTraceTest::test_compiled_autograd_attribution
• pull / linux-jammy-py3.10-clang18-asan / test (default, 2, 7, linux.4xlarge) (gh) (disabled by #156921 but the issue was closed recently and a rebase is needed to make it pass)
  test/dynamo/test_structured_trace.py::StructuredTraceTest::test_compiled_autograd_attribution
• pull / linux-jammy-py3.10-gcc11 / test (default, 3, 5, linux.2xlarge) (gh) (disabled by #156921 but the issue was closed recently and a rebase is needed to make it pass)
  test/dynamo/test_structured_trace.py::StructuredTraceTest::test_compiled_autograd_attribution
• pull / linux-jammy-py3.13-clang12 / test (default, 3, 5, linux.4xlarge) (gh) (disabled by #156921 but the issue was closed recently and a rebase is needed to make it pass)
  test/dynamo/test_structured_trace.py::StructuredTraceTest::test_compiled_autograd_attribution

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

[janeyx99]

An alternative to this implementation is adding a new shim function that returns a DeviceHandle, and then converts the DeviceHandle to the device. My understanding is that this current impl returns a new Device object.

  /// Returns a `Tensor`'s device.
  inline Device device() const {
    return impl_->device();
  }

This implementation in TensorBase.h suggests that we should be adding a new shim instead to return the same instance. What do you think?

[mikaylagawarecki]

Is my understanding correct that that would mean the Tensor within libtorch owns the device object being used by the extension?

When we pass this across the libtorch/extension boundary, might the device need to outlive the tensor object in libtorch and would returning the DeviceHandle cause any issues?

[janeyx99]

I don't think passing it through the boundary should change lifetimes if we're just rerouting pointers.