Method for the computation of division operations, reciprocal, square root and inverse square root.

摘要

Method for computing division operations, reciprocal square root and square root reverse.; The method for computing double-precision floating point format of the indicated operations is usable in numerical processors and microprocessors. It is characterized by the use of a polynomial approximation minimax second order to get an initial estimate of the values ​​of mutual and inverse square root, and the subsequent realization of a single modified iteration of Goldschmidt. Performing a single iteration of Goldschmidt allows a considerable reduction in latency proposed over previous methods, provided that the total circuit area increase significantly method. two possible architectures have implemented our method: the first calculates division and reciprocal, with very low latency. The second, using the same hardware resources enables the calculation of the four functions mentioned.