Add top-level CMake kernels target

[GregoryComer]

Add a top level kernels CMake target, which includes all configured kernel libraries. I validate this change by building a simple runner using the executorch_kernels target and verifying that it was able to build and run MobileNet v2, with and without selective build. Once the changes land in ExecuTorch, this will be long-term validated in executorch-examples CI.

There are a few follow-ups needed, which I will handle shortly. I still need to add merging logic for the YAML files to generate a single selective library that includes quantized and LLM ops, when configured to build them.

This is done in the context of top-level CMake targets, tracked in #12293.

[GregoryComer]

Stack from ghstack (oldest at bottom):

• Update docs to use top-level CMake targets #12807
• -> Add top-level CMake kernels target #12806

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/12806

• 📄 Preview Python docs built from this PR

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

❌ 1 New Failure

As of commit 42c53a6 with merge base ce9da63 ():

NEW FAILURE - The following job has failed:

• Build Presets / apple (ios) / build (gh)
  RuntimeError: Command bash /Users/runner/work/_temp/exec_script failed with exit code 1

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

[larryliu0820]

Nice

[larryliu0820]

I mean in the future it should be composable. I'm OK with having this for now.

[BujSet]

@GregoryComer Should this also check for a conflict with the OPS_FROM_MODEL API?

[BujSet]

Oh I see it there, later down.

[larryliu0820]

Thank you for adding this!

[larryliu0820]

@GregoryComer can you remove these options: https://github.com/pytorch/executorch/blob/main/examples/selective_build/CMakeLists.txt#L55 and let it use the ones you added

[lucylq]

Thanks for adding this to top-level cmake!

So, executorch_selected_kernels is a selective build library for portable_kernels only it seems? If a user wants to selectively build on optimized or custom kernel libs, they should either:

1. Update here (for test purposes, I guess).
2. Follow the current flow and create their own selective build lib.

[GregoryComer]

Yeah, as a follow-up, I intend to enable optimized (and portable_optimized) kernels. To enable selective build for other libraries (quantized ops, llm ops, etc.), I should just be able to use the merge_yaml functionality to generate a combined op yaml, right?

For user's custom op libs, I think they'll probably need to define their own targets, as we can't access their targets from here (unless we re-wrote as a macro/function, but we can look at that separately).

Backend op libs are another case that needs more thinking through, as well. But hopefully this should be a good starting point.

[GregoryComer]

@GregoryComer can you remove these options: https://github.com/pytorch/executorch/blob/main/examples/selective_build/CMakeLists.txt#L55 and let it use the ones you added

I've created a follow-up task to track this: #12948. Will do this shortly.