i see many many const values for bitmasking in math.d - wouldn't it be better to reduce this magic values down to float traits (and start using them)?