I was wondering if numpy does this because that seems obvious when working with array operations anyway but seems partially? https://numpy.org/doc/stable/reference/simd/index.html

I support knowing SIMD, but I think it’s worth mentioning that with good design depending on language and compiler you can get SIMD for free by auto vectorization.

At least Rust does this, and let’s you get away not touching explicit SIMD at all. It’s of course more fragile since you don’t always know if it’s done or not, and they can be skipped for minor changes in code and a, but with good loop hygiene you’ll get a lot of it for free.

For truly fast implementations though, handmade SIMDis usually better and faster if done correctly. And can do much more than auto vectorization can.