A "fun" thing to do if you really want to get a hang of IEEE floats is manually write the algorithms for a bunch of rounding operations (ceil, floor, truncate, half to even, etc).