Use default overloads when calling custom ops

[Erik-Lundell]

If a node is created without specifying an overload, A OpOverloadPacket is created, rather than an OpOverload. This works in a GraphModule, but the OpOverloadPacket is not a valid operator type in the _EXIREdgeDialectVerifier, which means that Edge ExportedPrograms can't contain a GraphModule with such ops.

In short, specifying using the default overload seems to be the more correct way of calling a custom
operator.

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/7836

• 📄 Preview Python docs built from this PR

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

❌ 3 Cancelled Jobs, 1 Unrelated Failure

As of commit b9a2862 with merge base e78ed83 ():

CANCELLED JOBS - The following jobs were cancelled. Please retry:

• trunk / test-models-macos (cmake, emformer_transcribe, xnnpack-quantization-delegation, macos-m1-stable, 90) / macos-job (gh)
  ##[error]The operation was canceled.
• trunk / test-models-macos (cmake, llama2, xnnpack-delegation, macos-m1-stable, 90) / macos-job (gh)
  ##[error]The operation was canceled.
• trunk / test-models-macos (cmake, mv2, portable, macos-m1-stable, 90) / macos-job (gh)
  ##[error]The operation was canceled.

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / unittest / linux / linux-job (gh) (trunk failure)
  backends/xnnpack/test/ops/test_add.py::TestAdd::test_qs8_add3

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

[Erik-Lundell]

CI Failure unrelated.

[digantdesai]

IIUC

executorch/exir/verification/verifier.py

Line 68 in 31f35d9

return super().allowed_op_types() + (torch._ops.OpOverloadPacket,)

allows OpOverloadPacket for torch ops but not for custom ops? Is that the main problem?

If yes, this is OK.

[Erik-Lundell]

IIUC

executorch/exir/verification/verifier.py

Line 68 in 31f35d9

return super().allowed_op_types() + (torch._ops.OpOverloadPacket,)

allows OpOverloadPacket for torch ops but not for custom ops? Is that the main problem?

If yes, this is OK.

I think

executorch/exir/verification/verifier.py

Line 239 in 31f35d9

return super().allowed_op_types() + (EdgeOpOverload, types.FunctionType)

is the relevent line but yes, that is the main problem