floating-point arithmetic

A floating-point library providing correct directed rounding to the last bit, with a precision of 24 bits and 7 bits of exponent.