Re: As simple as it is now, I am still against motion 24.03...
On 2011-06-10 02:57:30 -0400, Michel Hack wrote:
> Vincent Lef?vre wrote:
> > On 2011-06-09 03:52:28 -0700, Dan Zuras Intervals wrote:
> > > So I will vote against motion 24.03 & urge others to do so
> > > as well.
> >
> > I agree with Dan. Motions should not be on how interval arithmetic
> > should be implemented (internally), but on what it provides (the
> > behavior). Implementation details could be part of the rationale,
> > but not the subject of a motion.
>
> I don't see this motion as saying ANYTHING about implementation -- nor
> is there any mention anymore of the form of support for arithmetic or
> conversion operations with explicit directed rounding.
The text of the motion is:
Every IEEE 1788 compliant system shall provide the four basic
arithmetic operations addition, subtraction, multiplication,
and division with rounding downwards and upwards. Type conversions
with directed roundings shall also be provided.
It seems that the only goal of such a requirement is for a particular
implementation, as said in the introduction: "On the computer the
lower bound of the result of an interval operation is computed with
rounding downwards and the upper bound with rounding upwards." But
from my reading of the document, it doesn't bring anything for the
user of the standard.
> In an Interval Arithmetic context I would use the environment's IA
> operations for certain things, but would ALSO use directed-rounding point
> operations on others where I understand issues of local monotonicity, of
> points of discontinuity which I might avoid explcitly, and of evenness
> considerations. The latter is the reason why I find rounding towards zero
> (RTZ) useful too; in fact, I'd like RAZ (round away from zero) too (and I
> have that option for DFP on IBM's P and Z series).
So, you mean that a user of interval arithmetic may use either
interval operations or directed-rounding point operations, depending
on the context? But then,
1. Why the 4 basic operations only? If directed-rounding point
operations are necessary/useful, why not all the functions
supported by the 1788 standard?
2. Should these two kinds of operations/functions be covered in
one standard (more work) or two (there's already the IEEE 754
standard)?
Note that the motion doesn't say what 0 /> 0 is, for instance.
So, there's at least a big hole in it... unless one is in the
context of the IEEE 754 standard, but then, the motion becomes
useless. Or does the motion try to reinvent the IEEE 754 standard
(possibly a part of it) in its own way?
--
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 / Arénaire project (LIP, ENS-Lyon)