Enigma is a Python-first GPU compute framework for Apple Silicon. You write kernels in Python using a CuTe-inspired layout algebra, and Enigma compiles them to Metal Shading Language (MSL) and dispatches them through the Metal runtime.
What you need
- macOS with an Apple Silicon GPU
- Python 3.11, 3.12, or 3.13
- Xcode Command Line Tools (for
xcrun metal and xcrun metallib)
Python distributions above 3.13 are not supported.
In this section
| Page | What you will learn |
|---|
| Installation | Install Enigma, verify toolchain |
| First Kernel | Write, compile, and dispatch a vector-add kernel |
| Compile-Only Workflow | Iterate on codegen without a GPU |
The thirty-second version
import enigma
import numpy as np
@enigma.kernel
def add(A: enigma.f32, B: enigma.f32, C: enigma.f32):
tid = enigma.thread_position_in_grid
C[tid] = A[tid] + B[tid]
compiled = enigma.compile(add)
rt = enigma.MetalRuntime()
n = 4096
a = np.random.randn(n).astype(np.float32)
b = np.random.randn(n).astype(np.float32)
raw = rt.execute(compiled, [a, b], output_size=n * 4, grid=(n, 1, 1), threads=(256, 1, 1))
out = np.frombuffer(raw, dtype=np.float32)
np.testing.assert_allclose(out, a + b, rtol=1e-5)
print("ok")
