The Geode is currently configured for imprecise floating point exceptions. Is that what we want? (Imprecise exceptions make the FPU run faster, at the expense of exceptions potentially happening after the instruction that caused them. Precise exceptions limit the FP pipeline to one instruction at a time.)