The same way constant folding is done in any compiler. You try to emulate the math on the target.

D is quite specific what computations occur at compile time and what at runtime. It's very much up to the user in how he sets it up.