feat: support metal cpp

[drbh]

This PR supersedes #291

---

This PR adds support for a new metal-cpp kernel dependency. This is a follow up to the metal-cpp support in hf nix: huggingface/hf-nix#128 and enables kernels to use the cpp headers to drive metal kernels.

Changes:

• adds dep to build2cmake
• adds new relu-metal-cpp example
• builds example in CI

example usage

cd examples/relu-metal-cpp
nix build -L .
cd ...
uv run test_relu_metal_cpp.py

test_relu_metal_cpp.py

# /// script
# requires-python = ">=3.10"
# dependencies = ["kernels", "torch", "numpy"]
# ///
from kernels import get_local_kernel
import torch
from pathlib import Path

relu = get_local_kernel(Path("examples/relu-metal-cpp/result"), "relu").relu

input = torch.tensor([-1.0, -1.5, 0.0, 2.0, 3.5], device="mps", dtype=torch.float16)
out = relu(input)
ref = torch.relu(input)

assert torch.allclose(out, ref), f"Float16 failed: {out} != {ref}"

print(out.cpu().numpy())
print(ref.cpu().numpy())

print("PASS")

output

[0.  0.  0.  2.  3.5]
[0.  0.  0.  2.  3.5]
PASS

[danieldk]

Really cool!

Added some comments. Maybe you can also copy over the test from the relu example?

[danieldk]

Do we need a version like cutlass? My first guess is not, since on Mac we always have everything the latest, but I thought I'd check.

[drbh]

good question! I agree we probably do not need a version here and we'll always prefer latest

[danieldk]

Would be nicer to have these in a directory, since it's an example, we want best practices :).

[drbh]

totally agree, i've updated the folder struct and build.toml in latest

[danieldk]

Remove.

[drbh]

removed

[danieldk]

One more small comment (file to remove), CI should work again after a rebase on main.

[danieldk]

I think this is another development file that can be removed? (I think I missed it the first time.)