Comment by slonopotamus

I believe that could be optimized in the future. At least, by splitting amd64/arm64 code.