The implementation supports both the single precision (32-bit) and double precision (64 bit) formats described in the IEEE-754 standard. The ansif library only supports single precision, because double precision numbers are restricted to 32-bits for performance purposes. 80-bit extended precision is not supported.