Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: Fwd: Re: IEEE-SA Daily email notification - You have 1 new notice(s).



Neither IEEE 1788 nor IEEE 754 consider infinities as numbers.

In IEEE 754 the values inf and -inf denote the result of an overflow. In IEEE 1788 these special values are used to denote boundaries of an unbound interval -- not numbers.

For example, consider the following operation:

isMember (inf,[entire]) returns false according to IEEE 1788, where inf is a floating-point value according to IEEE 754.

So, I don't understand what you disagree with.

Oliver


Von: ulrich.kulisch@xxxxxxx
Gesendet: 1. August 2017 12:38 nachm.
An: stds-1788@xxxxxxxx
Betreff: Fwd: Re: IEEE-SA Daily email notification - You have 1 new notice(s).

I forward a recent e-mail exchange:


-------- Weitergeleitete Nachricht --------
Betreff: Re: IEEE-SA Daily email notification - You have 1 new notice(s).
Datum: Fri, 21 Jul 2017 10:18:17 +0200
Von: Ulrich Kulisch <ulrich.kulisch@xxxxxxx>
An: IEEE-SA <sa-ballot@xxxxxxxx>


Thank you for inviting me repeatedly to join the IEEE P1788.1 committee.

In the early days of the standard development I prepared several motions 
which all have been accepted. Among these motions was one that required 
an exact dot product (EDP).

Hardware implementations of the EDP at Karlsruhe in 1993 and at Berkeley 
in 2013 show that it can be computed in about 1/6th of the time needed 
for computing a possibly wrong result in conventional floating-point 
arithmetic. However in 2013, the requirement of an EDP was withdrawn 
from the standard, perhaps in order to make Sun's Forte Fortran conform 
with the standard. A majority of IEEE1788 members was made to believe 
that everything can be done as well by a correctly rounded dot product 
which is built upon a conventional computation of the dot product in 
floating-point arithemtic followed by an accuracy refinement. So it is 
slower than the EDP by more than one magnitude.

Another topic where I disagree with the majority of the IEEE1788 members 
is the treatment of the infinities. Interval airthmetic developed for 
closed real and floating-point intervals leads to an exception-free 
calculus (for proof see my books). IEEE1788, however, develops interval 
arithmetic over the IEEE754 numbers where ±infinity are considered as 
numbers. This pulls several IEEE754 concepts like NaN, +0, -0, NaI, and 
others into interval arithmetic where they are not needed and are a 
frequent source of confusion.

I attach a short unpublished paper which shows my view of the matter in 
more detail.

Best regards
Ulrich Kulisch





Am 11.07.2017 um 06:01 schrieb IEEE-SA:
> You have new myProject notifications - please log onto https://development.standards.ieee.org/my-site to read them.
>
>    ----Message Boundary----
>
>
> To: undisclosed-recipients:;
> Cc: "C/MSC Project Staff Liaison" <c.bahn@xxxxxxxx>
> Subject: Invitation for IEEE P1788.1 did NOT close! Your participation is important!
>
>
> This notification is to advise you that the invitation for P1788.1 Standard for Interval Arithmetic (Simplified) has been extended by the chair.  Please consider joining this ballot group and submitting your vote and comments on the document.  Log on to myProject https://development.standards.ieee.org/my-site/open-ballot-invitations "Show/Join Open Ballot Invitations" to find the ballot group.  Follow the on-screen instructions.  Your participation is vital to the success of this project.
>
> Thank you.
> IEEE-SA Balloting Center
>
>


-- 
Karlsruher Institut für Technologie (KIT)
Institut für Angewandte und Numerische Mathematik
D-76128 Karlsruhe, Germany
Prof. Ulrich Kulisch

Telefon: +49 721 608-42680
Fax: +49 721 608-46679
E-Mail: ulrich.kulisch@xxxxxxx
www.kit.edu
www.math.kit.edu/ianm2/~kulisch/

KIT - Universität des Landes Baden-Württemberg
und nationales Großforschungszentrum in der
Helmholtz-Gesellschaft