Dynamic Exponents for Floating Point Numbers-2022-04-07.pdf

Dynamic Exponents implement an improvement of IEEE 754 floating point formats with dynamic exponent size. Unused exponent bits are used to increase the number precision.

At the same time artificial fixed exponent size is replaced by an exponent size determined from the total type size only. The exponent can span up to half the total size, but can be restricted to 32 bits which outperform even the 256 bit IEEE format with only 19 bit exponents.

A simple combinative conversion hardware into IEEE 754 format is described, usable to extend the loadable IEEE data formats.