Add a MacOS builder

[rengolin]

After #267 Lighthouse is supposed to run on MacOS, but we need a CI builder to make sure it works consistently and doesn't regress.

I've added macos-26 to the matrix in https://github.com/llvm/lighthouse/tree/macos, but the error in Torch means we'll have to also update the toml file somehow.

error: Distribution `torch==2.13.0+cpu @ registry+[https://download.pytorch.org/whl`](https://download.pytorch.org/whl%60) can't be installed because it doesn't have a source distribution or wheel for the current platform
hint: You're on macOS (`macosx_26_0_arm64`), but `torch` (v2.13.0+cpu) only has wheels for the following platforms: `manylinux_2_28_aarch64`, `manylinux_2_28_x86_64`, `linux_s390x`, `win_amd64`, `win_arm64`; consider adding "sys_platform == 'darwin' and platform_machine == 'arm64'" to `tool.uv.required-environments` to ensure uv resolves to a version with compatible wheels
Error: Process completed with exit code 2.

Torch-MLIR also doesn't have support for MacOS (only manylinux and win), so that's another thing that we'll need to fix before we can officially claim to support MacOS.

[FedericoBruzzone]

Hi @rengolin, thanks for iterating on this🙏

I'm able to reproduce it locally. In my first setup, I deliberately skipped: uv sync --extra ingress-torch-cpu --extra runtime_mpich.
However, while for Torch, it's actually a simple fix since the +cpu postfix (within the toml file) is not needed on Darwin, for torch-mlir we don't have the macOS support right now (see here), as you pointed out :(

So, the torch-mlir-release README explicitly say:

At the moment only Linux (x86_64 + AArch64) and Windows (x86_64) builds are published. If you are interested in other configurations, please send a PR.

What do you suggest we do?

[rengolin]

What do you suggest we do?

Send a PR? 😄

Now seriously, it should be simple to do get macOS on the matrix, but getting all the tests to pass is a whole different matter. I think it would be an interesting conversation at the very least. Starting with an issue, and perhaps a mock PR that shows what crashes would be a good start.

[FedericoBruzzone]

Send a PR? 😄

I hadn't thought about that, you know? :'D

Now seriously, it should be simple to do get macOS on the matrix, but getting all the tests to pass is a whole different matter. I think it would be an interesting conversation at the very least. Starting with an issue, and perhaps a mock PR that shows what crashes would be a good start.

I'm working on this. I'll send a PR ASAP. Thank you!
I have an AI-assisted commit in which I'm running the GH action against (it works locally).