Browse Prior Art Database

Mehtod to Produce, Divide and Square Root Results with User-Defined Accuracy

IP.com Disclosure Number: IPCOM000115414D
Original Publication Date: 1995-Apr-01
Included in the Prior Art Database: 2005-Mar-30
Document File: 2 page(s) / 53K

Publishing Venue

IBM

Related People

Matheny, DT: AUTHOR [+2]

Abstract

Most graphic applications require a large number of divide and square root operations. Rarely does the accuracy of these operations require IEEE double or even single precision accuracy. This invention allows a user to select the accuracy of a divide or square root operation, with less accurate results requiring less cycles to produce.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 57% of the total text.

Mehtod to Produce, Divide and Square Root Results with User-Defined
Accuracy

      Most graphic applications require a large number of divide and
square root operations.  Rarely does the accuracy of these operations
require IEEE double or even single precision accuracy.  This
invention allows a user to select the accuracy of a divide or square
root operation, with less accurate results requiring less cycles to
produce.

      With the advent of today's integrated high-speed floating point
units, more and more applications are taking advantage of their ease
of use.  However, this has exposed a weakness as pertains to
instruction execution time for divide and square root functions.
These instructionsusually are performed in a dedicated hardware
section or by use of iterative techniques in the main pipeline of the
floating point unit.  In any case, these instructions take some
multiple of the time to perform a simple instruction, such as an add.

      Floating point units adhering to IEEE arithmetic standards are
required to produce the results of divide and square root in the
correctly rounded format assuming an infinite precision before
rounding.  This result can either be in single precision or double
precision format.  However, there are many applications (such as
graphics) where this amount of accuracy is not needed.  Below, a
method is presented to produce a result with less accuracy while
decreasing the time needed to produce that result.

     ...