(j3.2006) Interval arithmetic

[Keith Bierman]

On Thu, Mar 12, 2009 at 4:30 PM, Bill Long <longb at cray.com> wrote:
> ?I strongly suspect any intrinsic
> type spec would at least initially be implemented as a derived type
> internally, and thus there would be no difference at all.

If "sharpness" is important, the killer issue remains (as I mentioned
at the time) having user level code try to inform the optimizer what
can and can't be done. Since overly wide results are what made
previous attempts at interval arithmetic functionally useless in
practice (combination of using "point" algorithms along with naive or
no optimization) I don't think this is just "an implementation
detail".

Interval methods, to be useful, have to be reasonably fast (viz. as
fast as they can, since they are at such a disadvantage vis a vis
point methods) and reasonably sharp (or the merit is minimal. The
answer to anything is in the interval [-inf,+inf] making a processor
simple if sharpness isn't a strong goal ;>

"Legacy" intel chips are still a dominant life form in computing life.
I don't know that they shouldn't be a consideration; but I concur they
shouldn't be overweighted in such decision making. x64 processors tend
to closer to "pure" 64-bit machines than the x86 processors were (the
80-bit, 64-bit stuff, etc.).

-- 
Keith Bierman
khbkhb at gmail.com
kbiermank AIM