I believe the hardware version traps to the OS for the aread instruction which then executes a bunch of low-level instructions to do IO and parsing. You could

I believe the hardware version traps to the OS for the aread instruction which then executes a bunch of low-level instructions to do IO and parsing. You could start by making a CPU that traps to the OS for every single instruction (except for some minimal core), and then incrementally implement hardware versions to speed up execution.