Re: modal interval
Arnold Neumaier wrote:
> Apply the same ingenuity that you applied to the modal processor to
> a directed rounding processor, and you get a more flexible tool at
> a very similar hardware cost.
I'm glad you admit it was ingenious (thank you).
>
>> It is also worth mentioning that the paired processor approach is
>> not truly an "interval processor." IMHO, if one goal of 1788 is to
>> provide incentive for hardware vendors to manufacture interval
>> processors, this route is counter-productive to that goal.
>
> A processor with two pipelines for directed rounding in each direction
> can easily be adapted to allow both directed rounding calculations and
> an interval arithmetic (which consists mostly of the latter). The
> resulting processer is not more complex than a modal processor.
However, fact remains it does not handle important cases where branching
occurs (as in the linearInt() formula). In this respect, modal interval
arithmetic already provides the efficient hardware implementation and as
well as a general-purpose solution that has wide applicability and a larger
potential user base. Directed pipelines extention could be added to a modal
interval processor at little cost to give even better results and a wider
audience of customers. So I think these are reasons why 1788 should at least
strive to provide compatibility with modal intervals. Better yet, they would
be included. All IMHO, of course...
Nate