Browse Prior Art Database

Normalized Integer Operations for a Floating Point Arithmetic Unit

IP.com Disclosure Number: IPCOM000092575D
Original Publication Date: 1966-Dec-01
Included in the Prior Art Database: 2005-Mar-05
Document File: 2 page(s) / 38K

Publishing Venue

IBM

Related People

Frieman, CV: AUTHOR [+4]

Abstract

In scientific computation it is sometimes desirable to perform integer calculations on the integer value of floating point numbers. The drawing shows devices for providing this capability in an otherwise standard floating point arithmetic unit.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 63% of the total text.

Page 1 of 2

Normalized Integer Operations for a Floating Point Arithmetic Unit

In scientific computation it is sometimes desirable to perform integer calculations on the integer value of floating point numbers. The drawing shows devices for providing this capability in an otherwise standard floating point arithmetic unit.

For example, the data word format is assumed to be one bit fraction sign, one bit exponent sign, three bit exponent, and six bit fraction.

Integer quantities are represented as normalized floating point numbers as in this table.

(Image Omitted)

Additional instructions needed to work with these numbers are set integer value, integer add, integer subtract, and integer multiply.

Instruction set integer value converts either normalized or unnormalized floating point numbers to their normalized integer equivalent as shown. The drawing shows devices for implementing execution of this instruction.

The fraction sign does not change. The fraction magnitude is shifted right by the amount 6 - Exp for exponents in the range from 1 through 5, otherwise it is not shifted, to eliminate all bits to the right of the true binary point. It is then normalized by a left-shift equal to the number of leading zeros. The output exponent is equal to the input exponent plus the amount of right-shift minus the amount of left-shift. However, if the input exponent is less than 1 or if all fraction bits are lost in the right-shift operation, then the output answer is set to the represent...