Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Fast Execution of Single and Double Precision Operands in the Binary Floating Point Architecture

IP.com Disclosure Number: IPCOM000059911D
Original Publication Date: 1986-Feb-01
Included in the Prior Art Database: 2005-Mar-08
Document File: 1 page(s) / 12K

Publishing Venue

IBM

Related People

Rodriguez, JR: AUTHOR

Abstract

A technique is described whereby execution time of binary floating- point operations is optimized when the operand fraction has been rounded to single or double precision. In binary floating point operations, computation is usually processed in double extended format. This requires that single or double precision operands first be converted to double extended format. Although the use of double extended format, in the execution of intermediate operations, is beneficial in the numerical characteristics of arithmetical computations, the floating-point processing performance time is significantly reduced.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 66% of the total text.

Page 1 of 1

Fast Execution of Single and Double Precision Operands in the Binary Floating Point Architecture

A technique is described whereby execution time of binary floating- point operations is optimized when the operand fraction has been rounded to single or double precision. In binary floating point operations, computation is usually processed in double extended format. This requires that single or double precision operands first be converted to double extended format. Although the use of double extended format, in the execution of intermediate operations, is beneficial in the numerical characteristics of arithmetical computations, the floating-point processing performance time is significantly reduced. This is due to the fact that most multiply and divide operations are implemented based on multi-cycle algorithms, which require longer execution time for data in double extended precision than for data in either single or double precision. The concept described herein explores two architectural areas whereby floating-point processing time can be decreased. During the execution of an arithmetic operation, skip-over zero's methods should be used whenever possible. This method will be of benefit whenever the original operands appeared in single or double precision formats. That being the case, the conversion made to place the operands in the double extended format for execution by an arithmetic operation did not change the number of low-order zeros. Following this approach,...