Addendum to ',X' bug

[Joseph Comfort]

Tobias,

Tobias.Schlueter@Physik.Uni-Muenchen.DE wrote:
> Quoting Joseph Comfort <Joseph.Comfort@asu.edu>:
>> I have a collection of old nuclear physics programs, from the 70's and
>> 80's, and am checking them out.  They were coded mostly in Fortran IV,
>> with small changes later for IBM, VAX, and g77.  Apart from the 'nX'
>> bug, they all compile and run fine on i586 and x86_64 machines.  There
>> are some small numerical differences between the results from g77 and
>> gfortran, espeically in how numbers are rounded on the outputs.  The
> 
> Thanks for doing this.
> 
>> main 'problem' I see now is that compilation and execution times are
>> larger by as much as 50% or factors of 2-3 (all with -O2 optimization).
> 
> If you could isolate small testcases or determine where the losses lie, we would
> be very interersted in knowing more about these.  We hope that gfortran will
> soon be able to replace g77 (die g77, die!), and performance problems are
> definitely something that will hinder the transition from g77 to gfortran.
> 
That could be tricky, as the 'lengthy' parts might be down in the bowels 
after input and setup. I'll see what I can do to isolate something. 
Although there are significant time differences, it is also true that 
stacks of cases that once took a good fraction of an hour in the 'old' 
days (and costing money) now run in a fraction of a second.

>> The same story also holds for the alternative g95 compiler, except that
>> it has a couple of small (and known) bugs for x86_64 systems.
> 
> Does it have public bugtracking nowadays?
> 
I don't think so.  I just email Andy directly.  (He was once a graduate 
student in Physics at ASU.)

> - Tobi
> 

Joe
-- 
Joseph Comfort
Arizona State University