Method and device for fast calculation of the reciprocal of a decimal floating point number with a high degree of precision by use of look up memory tables and linear-iteration

摘要

Device comprises a processor and two memories configured as lookup tables for determining the most significant bits of the mantissa with outputs of a first array of significant bits and a second difference array. A formula is provided for determining the minimum size of a search table corresponding to the required precision as well as formulas for calculating rows of the search table. The search table comprises initial approximations and correction factors. A linear iteration method is carried out, possibly using a Newton-Raphson method. An Independent claim is made for a corresponding method for calculating the reciprocal of a normalized mantissa of an input floating point number.