Re: requiring hardware is futile
> David Hough 754R work schrieb:
>
>> Even so, the best way to get interval performance improved is probably
>> to get
>> 1788 finalized, then get language standards that support 1788, then get
>> realistic
>> open source applications written according to those language standards
>> that can be used to compare performance on
>> platforms, then get an organization like SPEC to select a few of those
>> to measure
>> interval performance, then get end users interested enough to make
>> procurement
>> decisions based on those results.
>
> I agree.
>
>
> Arnold Neumaier
>
In my opinion, dedicated hardware implementations can be considered only
by vendors. The purpose of standardization is the following: same program
with the same inputs running on two different processors must yield the
same result.It is very important to develop the standard, in order to
provide an unitary and clear methodology when dealing with the problems
that can occur in interval arithmetic. Thus, design decision regarding
issues in interval arithmetic can be made based on the standard. However,
regarding hardware implementations, only vendors can decide in this type
of problem. Hardware support meeting some performance expectation can
appear in general purpose processors, as most of us want (in this case
SPEC benchmarks are needed) . However, hardware support for some parts of
interval arithmetic may also appear in very dedicated hardware
accelerators of some algorithms (which require in some stages interval
computations). So, the decision to include hardware support is 90% made by
vendors.
Alexandru Amaricai