🚀 go-pugleaf

In article <1994Nov14.184755.786@guitar.demon.co.uk>
           marcus@guitar.demon.co.uk "Marcus Bainbridge" writes:

>Keith Pritchard wrote:
>
>> (1.0/824633702449.0)*824633702449.0 is not equal to 1.0!
>
>[not directed at Keith, but at whoever wrote the test program]
>
>If you're doing things like this to that many places, you need something
>better than 64-bit or 80-bit accuracy to guarantee that *all possible
>calculations* are accurate.  My Casio calculator says the result is 
>0.999999999.  Does this mean I can get them to replace the CPU?

All possible calculations should be correct to a particular degree of
accuracy determined (mostly) from the representation. 0.999999999 may be
adequate for the format your calculator uses but it is not accurate enough for
IEEE double precision format.

>OTOH, Windows calculator has trouble doing something like 1.02-1 (can't
>remember the exact sum, but the answer came out as 0 when it should have
>been 0.02).

In this case the problem can be shown by means other than Windows Calculator.
It sounds like quite a serious bug in Calculator if what you say is true
however (I assume this isn't a CPU bug).

-- 
-----------------------------------------
Lawrence Kirby | fred@genesis.demon.co.uk
Wilts, England | 70734.126@compuserve.com
-----------------------------------------