Did you skip the part where he told that 6 instructions are being executed in parallel per cycle or is this not tricky in your opinion?
# a += (*p++) * (*q++)
mov x qi , mov y pi , mov b p*q , add , mov a a+b , [fiddle with index registers or store result]
It only looks tricky because the operation is manually scheduled onto a fairly deep pipeline, so each unit is doing its phase of a different iteration of the loop in any given cycle.More ugh is something like managing delay slots which MIPS and other common chips used in the 1990s consoles used https://en.wikipedia.org/wiki/Delay_slot. POWER is a much nicer and post-modern ISA so I think the 2000s era consoles were a lot easier to develop for.