Throwing `-O9' at everything that would take it sped it up to 2:09, but the CPU was still pegged. Notably, a multiplication routine wouldn't take the -O9, and I'm working on that.

Edit: Fixed the two bugs that prevented `-O9' from working on those multiplication routines and it cut another 2 seconds off, but the CPU is still pegged. (Not surprising since they're hand coded ARM assembler anyway.)

Oh, and it's definitely the fixed point version.