Comment by iainmerrick

For performance, can’t you “just” have a swap-endianness instruction in your CPU, and have the compiler use it when it detects byte-shuffling code?

(That may even happen already on some architectures for all I know)