inf-sup,
inf-sup-nan,
neginf-sup-nan .
They are examples, not requirements for internal representation.
The only Level 3 representative of [0,0] mathematical interval in inf-sup representation
is a pair of (-0,+0).
There are 4 Level 3 representatives of [9,0] ininf-sup-nan, neginf-sup-nan representations:
pairs (-0,+0), (-0,-0), (+0,+0), (+0,-0).
Section 14.4 chooses the representation "inf-sup" as standard interchange format of bare intervals
with similar representation for decorated interval.
The requirement for interchange format is not a requirement for internal representation in memory.
Yes, unneeded calculation is necessary before IO into binary interchange format,
but not after any computational operation.
-Dima
----- Исходное сообщение -----
От: ianm@xxxxxxxxxx
Кому: stds-1788@xxxxxxxxxxxxxxxxx
Отправленные: Среда, 4 Декабрь 2013 г 23:28:15 GMT +04:00 Абу-Даби, Маскат
Тема: Re: Motion P1788/M0059: Level 3 description
The suggested requirement that the only representation of 0 be [-0, +0] would hurt performance compared to allowing both ranges to be either zero. Doing independent calculations for the two values will sometimes produce the other zero even with directed rounding, so after many operations there would have to be a check of each bound for the other zero to fix it, or an otherwise unneeded calculation to produce the desired zero.
- Ian McIntosh IBM Canada Lab Compiler Back End Support and Development
John Pryce ---2013-12-04 09:17:39 AM---P1788 On 2013 Nov 17, at 13:02, Ralph Baker Kearfott wrote:
| John Pryce <j.d.pryce@xxxxxxxxxx> |
| Ian McIntosh/Toronto/IBM@IBMCA, |
| 2013-12-04 09:17 AM |
| Motion P1788/M0059: Level 3 description |
P1788
On 2013 Nov 17, at 13:02, Ralph Baker Kearfott wrote:
> Since this motion has been made by Dmitry Nadezhin and seconded by
> Vladik Kreinovich, the discussion period herewith begins, and
> will continue until after the end of Sunday, December 1.
I haven't seen any discussion on Motion 59 except a comment Nov 18 by Michel Hack about a typo. Maybe that's why the Chair hasn't yet called for the vote to start.
Here are some comments.
1. §14.1 to 14.3 are about general principles: "What does representation mean?" etc. Should we say they shall apply to all flavors, hence move them to Chapter 1? (A few changes of wording would be needed. The shortened §14 would refer to them.)
2. End of 14.4 "-- choice of big- or little- endian byte order." Michel Hack, Nov 5, said endian-ness is more complicated than that.
If I understand right, eight bytes that we think are logically
7 6 5 4 3 2 1 0
might depending on the machine architecture be stored in any of the orders
3 2 1 0 7 6 5 4
5 4 7 6 1 0 3 2
6 7 4 5 2 3 0 1
I think this item should be replaced by, say, just "issues of endian-ness."
BTW the preceding items should be ended by semicolons not periods.
3. In 14.4 para 2, it mentions "the only representative of Empty is the pair (+oo, -oo)". I suggest add "... the only representative of [0,0] is the pair (-0, +0)".
John Pryce
