Browse Prior Art Database

Passing a Denormalized Number During Conversion of Floating-point Underflow Operations

IP.com Disclosure Number: IPCOM000060656D
Original Publication Date: 1986-Apr-01
Included in the Prior Art Database: 2005-Mar-09
Document File: 1 page(s) / 12K

Publishing Venue

IBM

Related People

Aggarwal, BK: AUTHOR [+3]

Abstract

A technique is described whereby denormalized numbers are considered in the architecture and design of binary floating-p conversion in computer architecture. Hardware and software considerations are discussed so as to pass a denormalized number to the trap-handler routine.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 72% of the total text.

Page 1 of 1

Passing a Denormalized Number During Conversion of Floating-point Underflow Operations

A technique is described whereby denormalized numbers are considered in the architecture and design of binary floating-p conversion in computer architecture. Hardware and software considerations are discussed so as to pass a denormalized number to the trap-handler routine.

IEEE standards set forth the necessary architectural standards in coping with denormalized numbers in the binary floating point conversion from a high precision number to one of lower precision. However, the standard does not spell out what the exponent adjust and destination format should be during underflow conditions in format conversions. The technique described herein outlines how to pass a denormalized number to the trap-handler routine using biased exponents and significand rounding to obtain the required precision.

In the computer's floating-point architecture, a trap condition for an overflow is set when an underflow-mask bit is set to one. When the mask bit is set, and an overflow occurs, the exponent of the resultant of the operands will be adjusted by some bias and the significand will be rounded to the desired precision and the final floating-point number will be sent to the trap-handlers. When the underflow exception mask bit is a one and the representation of a floating-point number is converted to a format of a lower precision, an exponent is returned. The significand is rounded to the precisi...