|Date: ||Mon, 4 Oct 1999 08:55:29 -0500|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|From: ||Michael S Hines <mshines@PURDUE.EDU>|
|Subject: ||Re: floating point lesson, please.|
|Content-Type: ||text/plain; charset="iso-8859-1"|
From the SAS Companion for the MVS Environment, Chapter 10
... All numeric values are stored in SAS data sets and are maintained
internally in floating-point (real binary) representation.
and later on....
... Like other numeric values, SAS maintains integer varialbes in
floating-point (real binary) representation. But under MVS, numeric integer
values are typically represented as binary (fixed point) values using two's
complement representation. SAS can read and write these values using
informats and formats, but it does not process them internally in this form.
I stand by my previous statement....
On the IBM SAS uses 8 byte real values which consist of one byte sign, seven
bits characteristic (bias of 040x), and 56 bits of mantissa (in hex). PP
141-143 explain the real binary representation and why rounding errors occur
do to approximations to representing numbers. The LENGTH statement can be
used to size these real numbers, but it affects precision - as the table on
pate 144 shows.
Michael S Hines, CISA,CIA,CFE,CDP
Coordinator of Data Systems / Programming
Schools of Engineering
1280 Engineering Administration
West Lafayette, IN 47907-1280
Phone 765.494.5338 Fax 765.496.1466
From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU]On Behalf Of F a
b r i z i o
Sent: Monday, October 04, 1999 6:33 AM
Subject: Re: floating point lesson, please.
Michael S Hines wrote in message ...
>SAS does not have a data type of integer... all numbers are floating
Though this is true, SAS is able to store any integer number
whose absolute value does not exceed 2**53 (for PC and
Unix environments) of 2**56 (for mainframe environments).
This exceeds the capacity of virtually any data type 'integer'
which usually does not accept numbers over 2**31-1 (apart
from some implementations of true 'decimal' hardware numbers
which accept up to 31 decimal digits).