g77 fortran bug ??

[Guilherme Regis]

Dear Toon Moene,

Thanks a lot for the information.
Would you please provide the directions to a document 
about single and double precision representations, exponent,
mantissa, etc, in g77 ?

Thanks again,
  Luiz

On Wed, 18 Sep 2002, Toon Moene wrote:

> Guilherme Regis wrote:
> 
> > Here is a simple test code called bug.F which
> > reads a number from a file and writes it on screen
> > and to an output file bug.out:
> > 
> >       program bug
> >       real*4 vec
> >       open(1,file='bug.dat',status='old')
> >       open(2,file='bug.out',status='new')
> >       read(1,*)vec
> >       write(2,*)vec
> >       write(6,*)' vec =',vec
> >       stop
> >       end
> > 
> > where bug.dat contains the line:
> > 
> > 	lgemedia:~$ more bug.dat
> > 	 839380840
> > 
> > The result is in bug.out:
> > 
> > 	lgemedia:~$ more bug.out
> > 	  839380864.
> 
> REAL*4 (or just REAL) in g77 is mapped on most target architectures to 
> IEEE-754 single precision floating point numbers.
> 
> The exponent range gives you numbers between 10**(-38) .. 10**38, but 
> that doesn't mean *arbitrary* numbers between those ...
> 
> IEEE-754 single has a 24-bit mantissa, which means that every number 
> thus represented has about 7 decimal digits of accuracy.
> 
> The above numbers differ in the 8th decimal, which is in good agreement 
> with the above observation.
> 
> Hope this helps,
> 
>