Re: Meeting the Scope and Purpose of P754

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

> two-sum idiom and the like to implement distill operations.  (At various
> points those were required as primitives, but we (the P754 WG) had trouble
> agreeing on the details, and they were finally dropped.)

To combine this point point with a previous one, the sticking point was
whether the exception behavior should be specified in detail or not.
And that reduced to a judgment about whether the instructions would be
implemented in hardware.

For the types of operations for which this discussion has arisen,
testing whether an exceptional condition exists is no big deal in hardware,
because it can usually done without delaying the normal case.
But in software the extra testing does have a cost that almost never pays
off, because exceptional operands are so rarely encountered by these
operations which usually handle prefiltered normal data.    So fully-specified
exception response (for tail functions, or logb for instance) guarantees
that the software implementations will seldom be used, in favor of something
quick and dirty that works fine for normal data.     But partially-specified
hardware instructions are just a prescription for future compatibility regrets.
So even though the standard can be implemented in any combination of
hardware and software,
how much specification to put into the standard does indeed involve a guess
about future hardware.