Ah, yes.
Actually, the relevant code is not at all complicated. Just look at miller_rabin_pass() in lines 59 to 93 of miller-rabin.c (on the page linked above).
Most of the other code there is just a simple big-integer library, so you can test primality on natural numbers much bigger than what the hardware would otherwise support.
Perhaps you didn't notice that the original poster talked about integers in the millions to tens of millions -- well within 32-bit integer range --, not integers with millions to tens of millions of digits?