Browse Prior Art Database

Loading of Single Precision Denormalized Number

IP.com Disclosure Number: IPCOM000103068D
Original Publication Date: 1990-Jun-01
Included in the Prior Art Database: 2005-Mar-17
Document File: 1 page(s) / 41K

Publishing Venue

IBM

Related People

Fry, RE: AUTHOR [+2]

Abstract

The Floating Point Unit is a high performance machine that executes floating point load every cycle. All internal data format of floating point is represented in the IEEE double precision number. This invention describes an implementation that allows the loading of single precision denormalized number from a storage location to the floating point register without adding extra cycles or hardware to pre-normalize the number during a load single instructions. - The floating point load single instruction converts the single precision denormalized number into a double precision un-normalized number with the double precision un-normalized number tag set to one. On subsequent arithmetic operation, control logic can look at this bit and decide whether or not pre-normalization is needed.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 95% of the total text.

Loading of Single Precision Denormalized Number

      The Floating Point Unit is a high performance machine that
executes floating point load every cycle.  All internal data format
of floating point is represented in the IEEE double precision number.
This invention describes an implementation that allows the loading of
single precision denormalized number from a storage location to the
floating point register without adding extra cycles or hardware to
pre-normalize the number during a load single instructions.
   -  The floating point load single instruction converts the single
precision denormalized number into a double precision un-normalized
number with the double precision un-normalized number tag set to one.
On subsequent arithmetic operation, control logic can look at this
bit and decide whether or not pre-normalization is needed.  The
hardware for pre-normalization is the same hardware that executes the
floating point arithmetic instruction.
   -  Store-Double instruction has to pre-normalize the number if the
double precision un-normalized tag bit is set before writing it back
to memory.  No extra hardware is needed since this can be treated as
a pre-normalization step of regular arithmetic instruction.
   -  Store-Single instruction converts the data into single
precision format without any pre-normalization or denormalization
process.  No pre-normalization or denormalization is required because
an un-normalized double precision is the same as a de...