Browse Prior Art Database

Unnormalized Number Handling in a Floating-Point Unit

IP.com Disclosure Number: IPCOM000036423D
Original Publication Date: 1989-Sep-01
Included in the Prior Art Database: 2005-Jan-29
Document File: 2 page(s) / 56K

Publishing Venue

IBM

Related People

Cocanougher, D: AUTHOR [+2]

Abstract

High speed IEEE standard 754 Floating-Point Units (FPUs) must evaluate instructions with very few logic levels. Low cost machines must use smaller numbers of circuits. In both instances the normalization of all numbers from the multiply and add portion of the logic can require large shifters. This invention reduces the number of circuits while also reducing the number of levels required for normalization. The following technique is used: 1. During the full add operation the shift count is created by the Leading Zero Anticipator logic which also reports to the control whether the number can be normalized or not. 2. If the number can not be normalized, the shifter normalizes the number the maximum shift count allowed by the hardware.

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 86% of the total text.

Page 1 of 2

Unnormalized Number Handling in a Floating-Point Unit

High speed IEEE standard 754 Floating-Point Units (FPUs) must evaluate instructions with very few logic levels. Low cost machines must use smaller numbers of circuits. In both instances the normalization of all numbers from the multiply and add portion of the logic can require large shifters. This invention reduces the number of circuits while also reducing the number of levels required for normalization. The following technique is used: 1. During the full add operation the shift count is

created by the Leading Zero Anticipator logic

which also reports to the control whether the

number can be normalized or not.

2. If the number can not be normalized, the shifter

normalizes the number the maximum shift count

allowed by the hardware. The control logic

decides to place the unnormalized number into the

Register File so that later it can be normalized.

It does not allow the unnormalized number to be

used in arithmetic operations.

(Image Omitted)

3. A tag is placed with the unnormalized number when

placed in the Register File.

4. When the operand is read from the register file to

be used in new arithmetic operations or to be

stored, the tag tells the control to first

normalize the number before it is used. This

normalized number will replace the old

unnormalized number in the Register File so that

this inconvenience will only occur once.

Shifters and Leading Zero Anticipators which are used to normalize results are...