Browse Prior Art Database

FPU Special Case Logic Without MADD Array Dataflow

IP.com Disclosure Number: IPCOM000122211D
Original Publication Date: 1991-Nov-01
Included in the Prior Art Database: 2005-Apr-04
Document File: 2 page(s) / 70K

Publishing Venue

IBM

Related People

Chu, TV: AUTHOR [+4]

Abstract

The IEEE Standard for Binary Floating Point has some data formats which require special handling. These formats will not obey normal arithmetic operations. Manipulating these special formats in the MADD array dataflow will cost extra hardware and degrade performance by increasing the time required to process each cycle.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 56% of the total text.

FPU Special Case Logic Without MADD Array Dataflow

      The IEEE Standard for Binary Floating Point has some data
formats which require special handling.  These formats will not obey
normal arithmetic operations.  Manipulating these special formats in
the MADD array dataflow will cost extra hardware and degrade
performance by increasing the time required to process each cycle.

      The special formats can be categorized as:  NANs (Not a
Number), Infinity and Zero data.  Also included in the special cases
are operations which are invalid or require special treatment.  These
special numbers can be loaded to floating point or generated from the
arithmetic operations such as:  dividing a number by zero, dividing a
number of infinity, or executing arithmetic operations on these
special numbers.

      The solution is to detect these special cases at the decode
cycle, treat them totally in control logic, thereby avoiding the
dataflow.

      At the decode cycle, the operations which have special numbers
or will generate a special number in the course of arithmetic
operations are identified.

      The normal operations are directed to the MADD array (dataflow)
directly.  In cases where NAN-type operations are detected, the
decoder will route it to a MOVE-type operation, as shown in the
figure, by the output numbered 1 of the decoder.  Because the result
of any operation on NANs will generate a NAN, and since the result is
previously known, there is...