Re: [Help-gsl] Iterations number with Conjugate PR gradient minimization

[Mauro]

2008/1/31, Mauro <address@hidden>:
>
> 2008/1/30, Brian Gough <address@hidden>:
> >
> > At Mon, 28 Jan 2008 12:22:32 +0100,
> >
> > Maybe rounding.  Try using gsl_ieee_env_setup and
> > GSL_IEEE_MODE=double-precision,mask-all.
> >
> > --
> > Brian Gough
> >
>
> Hi, thank you for your answer!
> It's not IEEE rounding mode, I tried with some combinations between
> precision and rounding mode, but results don't change.
> But it looks a rounding problem, althoug I not understand if loss of
> precision is in memcpy (not probable) or in element-by-element copy (but
> would be a strange behavior)....
> Regards
> Mauro


Hi! I resolved the problem.
Valgrind creates a special environment, different from original, for execute
the program to be tested.
This environment coincides with IEEE double precision mode with round-up
rounding, and then executing the program with
GSL_IEEE_MODE=double-precision,round-up results are the same that I obtain
in Valgrind!
For differences between memcpy and element-by-element copy, there was a bug
in my code... :-\
Thank you for your support!
 Mauro