[cam-users] pgf90 compiler differences

Norm Wood nbwood@lamar.colostate.edu
Thu, 17 Jul 2003 10:52:24 -0600


Andy,

On 17 Jul., Andrew Mai wrote:
> Norm Wood wrote:
> 
> >No, I don't think so, but my expectations may be a bit naive.  In many
> >cases, the differences I'm seeing are in the 6th - 7th decimal place.
> >That's several decimal places worse than I'd expected with double
> >precision arithmetic, allowing for some progressive loss of
> >precision over the course of a calculation.  
> > 
> >
> I'm not sure what you are referring to here. Changing the *least* 
> significant bit in a single floating point number in the initial 
> conditions will change *all* of the numbers in the first or second 
> significant decimal place after a couple of weeks of integration.

Sorry - I wasn't being very clear with that "over the course of a calculation"
comment.  These differences in the 6th -7th decimal place occur in the 
results stored in the h0 history tape for the first timestep (time = 0.)

Norm

-- 
Norm Wood

     Department of Atmospheric Science
     Colorado State University
     norm@atmos.colostate.edu
     phone:  970-491-8587