Re: Request for motion [Fwd: Input from IFIP WG 2.5 to IEEE Interval Standards Working Group]
> From: John Pryce <j.d.pryce@xxxxxxxxxxxx>
> Subject: Re: Request for motion [Fwd: Input from IFIP WG 2.5 to IEEE Interval Standards Working Group]
> Date: Thu, 10 Sep 2009 18:42:15 +0100
> To: stds-1788@xxxxxxxxxxxxxxxxx
>
> Arnold
>
> On 10 Sep 2009, at 12:25, Arnold Neumaier wrote:
> > Dan Zuras Intervals wrote:
> >> If we are to pursue this at this time, please include accurate
> >> versions of all of sum, dot product, sum of squares, & sum of
> >> absolute values for all supported precisions.
> > ...
> > I think only interval-valued results for noninterval inputs should
> > be provided by the standard. Then there are no problems.
> >
> > If some element is NaN or two terms in the sum are +inf and -inf,
> > the result should be the empty set; otherwise the tightest enclosing
> > interval of the exact result should be returned.
>
> Am I missing something? Suppose we implement as you say, "interval-
> valued results for noninterval inputs", and just ONE term of the sum
> is, say, +inf, the rest being finite numbers. Then the result is
> +inf, but since P1788 is based on the reals, there is no interval
> that encloses this. What to do?
>
> John
Exactly, John. Even if some other term should be a NaN.
This is why I cautioned care in the definition of the behavior
of these reductions in exceptional cases. We took some time
on just these issues in clause 9.4 of 754-2008. There may be
some differences with interval applications but all the cases
for the floating-point functions are there in the text.
Again, there was also some controversy about our choices. It
is a valid criticism to observe that some choices make speed
or parallelism difficult to accomplish. We may wish to review
these issues for 1788.
Yours,
Dan