[thyrsus]

At what level should one code to be confident the backend will do proper chip-specific optimization? RTL? Or is that already too low level?