Apparatus and method for estimating shift amount when performing floating-point subtraction

Abstract

The disclosure relates to an apparatus and method for estimating a shift amount when performing floating-point subtraction. An apparatus and a method are provided for estimating a shift amount to perform a subtraction operation and generate a difference value. Shift estimation circuitry then determines an estimated shift amount to be applied to the difference value. The shift estimation circuitrycomprises significand analysis circuitry to generate a first bit string identifying a most significant bit position within the difference value that is predicted to have a bit set to a determined value according to analysis of the significand values of the two floating-point operands. Shift limiting circuitry generates a second bit string identifying a shift limit bit position from an exponent value. The shift limiting circuitry has computation circuitry to perform an associated computation using bits of the exponent value to determine a value for that bit position within the second bit stringfor each bit position in at least a subset of bit positions of the second bit string. Combining circuitry then generates a combined bit string from the first and second bit strings, and shift determination circuitry determines the estimated shift amount from the combined bit string.

Description

technical field [0001] The present disclosure relates to apparatus and methods for estimating shift amounts when performing floating point subtraction. Background technique [0002] It is common to use floating point (FP) representations in data processing systems. Floating-point numbers include a significand and an exponent indicating the significance of the bits of the significand. A normalized floating-point number has a significand of at least 1 and less than 2, so the format is 1.xxxxx. When subtracting one floating point number from another, the result can be very small numbers, i.e. there can be a large number of leading zeros. In seeking to renormalize the significand of the result, a left shift operation can be performed on the difference obtained by subtracting one significand from another. [0003] It is known to use a leading zero predictor circuit (LZA circuit) to predict how many leading zeros will be present in the difference based on an analysis of the sig...

AI Technical Summary

Problems solved by technology

Exponential decrement becomes a problem when the exponent is already small, since FP numbers cannot have a true exponent lower than a predetermined minimum

Therefore, any proposed left shift produced by LZA needs to be qualified so that no left shift that causes the exponent to decrement below the minimum value is applied

Images