Re: Motion M0033.01 -- Number format: YES
On 2012-05-29 15:50:18 -0400, Michel Hack wrote:
> I vote YES on Vincent's Motion 33.
>
> I have a mild uneasiness about the interpretation that 754-conforming
> implies that 1788 maintain the distinction between signed zeros. (B3)
Well, Motion 33 introduces a distinction, because it may be useful
for future motions (to be decided later).
This does not mean that functions will have to specify the sign of 0
(if it happens that the sign of 0 will never be specified, B3 would
become useless as a consequence). B3 also means that for each function
where this matters, whether the sign of 0 should be specified or not
needs to be discussed. One question that hasn't been answered yet is:
do we want full reproducibility for the 754 formats, knowing the fact
that 1/+0 and 1/-0 are different? If yes, then the sign of 0 will need
to be specified (possibly in an optional manner).
> Surely we know of no cases where the distinction would be meaningful
> in the P1788 context?
The functions concerned by the sign of 0 are those that will return
a number, i.e. you will leave the P1788 context for the 754 context.
Hence the possible need for such a distinction.
> Or rather, surely we DON'T WANT TO INTRODUCE such a distinction,
> e.g. width(singleton) = +0 but width(empty) = -0?
This should be discussed later. For instance, an application may
expect that (at least on non-empty intervals), 1/width(I) >= 0 on
Rbar. So, width(singleton) = +0 would be safer.
> I also think that a clarification is needed for rounding direction.
> The applicable rounding direction would be determined by the operation,
> and never by 754's concept of "current rounding direction", right?
That's a possible choice (already suggested in the previous
discussions). This hasn't been formally decided yet.
--
Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)