[Fwd: Formatted Read Accuracy]

[Fwd: Formatted Read Accuracy]

[Emil Block]

[-- Attachment #1: Type: text/plain, Size: 1 bytes --]

 

[-- Attachment #2: Type: message/rfc822, Size: 1095 bytes --]

From: Emil Block <blime@his.com>
To: kisa@centropolisfx.com
Subject: Re: Formatted Read Accuracy
Date: Wed, 09 Oct 2002 14:00:35 -0400
Message-ID: <3DA46EC3.F332AA1A@his.com>

Gokhan Kisacikoglu wrote:
> 
> > input xllt is  67.9936  and it becomes 67.9935989
> >
> 
> These happen because of floating point accuracy, it has nothing to do
> with the reader or your format statement. Certain numbers can not be
> accurately represented with float numbers (even with double
> precision)...
> 
> Good luck,
> -kisa

I don't understand -- the numbers  are correctly read with the F77 compiler 
used on the same source code.  This is only one example of hundreds of numbers
that are read and G77 adds change in the fifth thru seventh decimal places on 
all of them.

Blime

Re: [Fwd: Formatted Read Accuracy]

[Eljay Love-Jensen]

Hi Blime,

Floating point numbers -- IEEE 754, 32-bit "float", 64-bit "double", or any 
of the other more-or-less standard formats -- cannot perfectly represent 
any arbitrary decimal portion of all rational numbers.

For instance, 0.5 can be perfectly represented.

Consider that "0.5" (decimal) is "0.8" (hexadecimal).  Exactly.

0.1 cannot be perfectly represented.

Consider that "0.1" (decimal) would be 
"0.1999999999999999999999999999999..." (hexadecimal).  Since floating point 
numbers do not have infinite precision, some rounding occurs.

To alleviate the problem, consider using fixed point numbers, or using BCD 
numbers.

Sincerely,
--Eljay