Re: motioin43 amended
Vincent,
> > I assume that the use of -0 has been discussed and rejected some
> > time previously by 1788 committee.
>
> No, it was just decided that all intervals are closed. Then how
> they are represented (Level 3) or encoded (Level 4) is currently
> unspecified.
It is not specified how they are represented internally.
Section 14.3 of draft 7.1 says that representation of 0 in interchange format is +0 .
-Dima
----- Исходное сообщение -----
От: vincent@xxxxxxxxxx
Кому: stds-1788@xxxxxxxxxxxxxxxxx
Отправленные: Четверг, 25 Апрель 2013 г 3:51:50 GMT +04:00 Абу-Даби, Маскат
Тема: Re: motioin43 amended
On 2013-04-24 08:08:43 -0700, Richard Fateman wrote:
> On 4/24/2013 7:54 AM, Vincent Lefevre wrote:
> >On 2013-04-24 07:24:24 -0700, Richard Fateman wrote:
> >>I am hoping that to the largest extent the use of the IEEE-754
> >>infinities will be transparent to this distinction. I am perhaps
> >>naively assuming this comes up only in circumstances in which one
> >>computes 1/oo or 1/(-oo) and gets not-quite zero. Allowing the use
> >>of -0 provides an opportunity to encode [-oo,0) as [-oo,-0].
> > From what I've heard, this was done by some implementations and this
> >was the reason why sqrt(-0) was chosen to return -0. However I wonder
> >whether introducing some particular non-closed intervals would be
> >useful. And it would make the standard more complex.
> I think the only interval endpoints that would signify open would be
> +-oo and -0. I agree that -0 introduces a non-uniformity, but it
> slightly simplifies the programming, at least if 1/(-oo) -> -0
> automatically. The big advantage is that 1/(-0) comes out as -oo
> rather than oo. That is, I think, a selling point. Having
> open/closed endpoints at every real would be more complicated, yes..
I think that when generating a 0 as a left or right endpoint, the
computations wouldn't be more complex (but not simpler either),
since the rounding modes behave as expected, at least for the
arithmetic operations (+, -, *, /, sqrt). However considering both
kinds of intervals as inputs would make the code more complex.
> I assume that the use of -0 has been discussed and rejected some
> time previously by 1788 committee.
No, it was just decided that all intervals are closed. Then how
they are represented (Level 3) or encoded (Level 4) is currently
unspecified.
--
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)