zlacker

That is fantastic. Swift in the Mac has something similar thanks to the Accelerate framework. What I’ve seen, however, is that if you are targeting a very specific platform (for example, the M1 family), you get better performance by directly using the intrinsics supported by that platform. In the real world, you are unlikely to actually have such a specific target, so this may not be so important.

I look forward to seeing some benchmarks with .NET - Microsoft needs to support a pretty wide variety of platforms. It will be interesting to see if their implementation is better!

>>Kon-Pe+(OP)
Accelerate is a bit different in being both overall more high level API and specific to Apple (and abstracts away the usage of AMX and ANE too). On the other hand, Vector<T>/Vector128/256/512<T> in .NET is what is 'portable-simd' to Rust except it is not in preview and widely used across standard library where applicable.

As of now, Vector<T> automatically targets AVX2, SSE4.2 and AdvSimd (NEON). Vector256<T> targets AVX2 (for the lack of ARM counterpart) and Vector128<T> targets SSE / AdvSimd respectively.