Hello,
As I understand the issues, Interval Arithmetic has to be "part of
the language" because one needs precise control of rounding in
order to do a reasonable job of IA. Indeed, IMU that the subject
(mathematically) was immature for a standardization effort, and
in fact some original mathematics was published as a result of
J3 asking "OK, exactly how do we do this?". Also, "opaque types"
is an issue (meaning a type where one can't "peek inside" the
way one can with complex-as-two-reals).
That delayed the schedule for IA, and pushed it off f2k. The
"Interval Arithmetic Enabling Features" includes such things as
control of rounding on an op-by-op basis, control of rounding in
I/O formatting/unformatting, etc. I would hate to try all this
as a module, but I suppose YMMV.
Note that I was not a committee member when this was being discussed,
and perhaps one of the then committee members would like to comment
further (more precise control of rounding, er, ah, spin? :-).
Peter Shenkin wrote:
>
> Hi,
>
> Glenn, I'm sure others who know more about it will also respond, but
> no, interval arithmetic is not like fixed point. As I understand it,
> it is more like propagation of error through multiple computations,
> initiated by specifying the input variables as ranges, rather than
> single values.
>
> What I don't understand is why this has to be part of the standard.
> It would seem that one should be able to write modules that do
> this. What am I missing?
>
> -P.
>
> On Fri, 26 Nov 1999, Glenn Carver wrote:
>
> > At 9:43 pm +0000 22/11/99, Bill Moffitt wrote:
> >
> > Bill,
> >
> > >I believe that the implementation of intervals in Sun's new F90/F95
> > >compiler is a useful and pretty complete one that might well be emulated
> > >by others and eventually included in the standard. However, in this
> > >area, we certainly felt that practice should precede the standard so
> > >that implementation problems could be "wrung out" before
> > >standardization.
> >
> > I don't know very much about interval arithmetic but I'm assuming it's not
> > unlike fixed point arithmetic. I presume the advantage is that it's
> > significantly faster than normal floating point. Could you give us some
> > kind of indication of the speed increase that might be obtained?
> >
> > Glenn
> >
> >
> > ----
> > Dr. Glenn Carver, Senior Research Associate,
> > Centre for Atmospheric Science, Chemistry Dept., Cambridge University, UK
> > mailto:[log in to unmask] http://www.atm.ch.cam.ac.uk/~glenn/
> >
> > "I never think of the future, it comes soon enough"
> > - Albert Einstein
> >
> >
>
> --
> ** Whether the playing field is level depends on the coordinate system. ***
> ********* Peter S. Shenkin; Schrodinger, Inc.; (201)433-2014 x111 *********
> *********** [log in to unmask]; http://www.schrodinger.com ***********
--
Cheers!
Dan Nagle [log in to unmask]
Purple Sage Computing Solutions, Inc.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|