Hey, as long as we're talking about Go performance - can we please, please get some kind of wide vector intrinsics (ie, no cgo overhead) in a library, or at lea

Hey, as long as we're talking about Go performance - can we please, please get some kind of wide vector intrinsics (ie, no cgo overhead) in a library, or at least aggressive compiler generation of vector ops that actually use AVX & the ARM NEON equivalent? Right now peak floating-point performance isn't even within half of what it should be on a very recent CPU, and I'd love to be able to deploy Go that exposes machine learning models to a network interface.