Emelia/Emi on Nostr: Technically you wouldn't even need to add a NOP, since the first instruction would ...

Technically you wouldn't even need to add a NOP, since the first instruction would never actually be executed (unlike with delay slots), you can just subtract one instruction-length from the target address when assembling it (or simply define it as "the instruction *prior* to the target function, which need not be a valid instruction", similar to PDP-8 subroutines. Although that also wrote the current PC to the target address to function as the return pointer)