Re: differences between implementations of IEEE-754 basic operators

[David Hough 754R work <754r@xxxxxxxxxxx>]

> one can validly argue that it is silly to expect absolute 
> repeatable floating-point behavior

It might be unrealistic to expect it at the current time, but as a general
proposition one might well wonder why "standard" arithmetic is so unpredictable.
And many inexperienced people do so wonder.    I wouldn't call them silly.

Experienced people have been told that reproducible results are impossible
so many times that they may have come to believe 
irreproducibility to be preordained like
death and taxes.  And vendors have used that belief
to their advantage to cover a multitude of sins in hardware and software,
usually in the name of "optimization" and while there is a domain of
rather important computations for which reproducible results would not
be economical, the typical floating point computation (in Excel) is not in
that domain.

In a free market, presumably vendors will respond to customer demand over
time.    The interesting question is whether the 754R material on 
reproducibility, limited as it is, will provoke enough customer demand to
elicit a change in vendor behavior.

In contrast, 
another of the periodic waves of attached processor solutions is beginning
to crest.    Like previous waves, attached processors tend to be very
hard to program, tend to be a step away from standard toward nonstandard
arithmetic, and offer orders of magnitude performance improvement
for the right applications tackled by the right programmer -
but for the typical application and the typical programmer,
little benefit for a lot of work.
Maybe this wave will be different, but history is not encouraging.