Re: Motion P1788/M0004.01_Keep_to_754 NO
Dear George,
I don't understand your comments about the "we," "they" and "competitive
advantage." When I mention commercial potential, I just mean:
is it something I would want to use in my own interval products and
applications?
Siegfried's suggestion that 0*X=0 is a very nice property for intervals, but
if 1788 must be strictly 754-compliant, I must have 0*X=NaI when X is
unbounded. At Sunfish, we have opportunity to create our own interval
hardware for the computer graphics application. So we can do either option.
But why should we choose the option that leads to unwanted exceptional
conditions?
By almost any interpretation, 1788 will be breaking new ground regardless
what we come up with. It is not too different from the situation of the
first IEEE 754 standard, for example.
Sincerely,
Nate
P.S. It is worth noting that non-standard extentions to IEEE 754 is part of
"existing practice" that is also well-studied in the interval literature.
Even at Sunfish, we use a modal interval arithmetic with slight deviations
from IEEE 754 for about five years now, too.
----- Original Message -----
From: "Corliss, George" <george.corliss@xxxxxxxxxxxxx>
To: "Nate Hayes" <nh@xxxxxxxxxxxxxxxxx>; <STDS-
1788@xxxxxxxxxxxxxxxxx>
Sent: Thursday, May 07, 2009 2:13 PM
Subject: Re: Motion P1788/M0004.01_Keep_to_754 NO
I heard an interesting talk a couple days ago by a grizzled old veteran of
the standards wars in the area of industrial automation. He spoke on "Using
the Standards Process for Competitive Advantage." His view favors the
"codify existing practice." When standards try to break new ground, they
NEVER are implemented, in his view. Interesting academic exercises,
perhaps, but of no use.
Who thinks they NEED an interval arithmetic standard?
I suspect for most of us, "we" think "they" need an interval arithmetic
standard, but "they" don't see the need.
George
On 5/7/09 1:40 PM, "Nate Hayes" <nh@xxxxxxxxxxxxxxxxx> wrote:
> Nate and stds-1788,
>
> Ah, yes. You have touched on something at least
> some standards bodies have
> been grappling with for decades: Do standards organizations
> codify existing practice, do they develop and mandate newer
> and better ways, or do they do both?
>
> Best regards,
>
> Baker
I believe it is an important question for the 1788 working group to find
agreement on.
On the one hand, it makes our job easier to simply standardize existing
754-compliant practices. On the other hand, I don't see this approach
maximizes the full commercial potential of interval computations.
IEEE 754 is one of the most successful computing standards, ever. So clearly
it is good idea to build on that reputation and legacy. But if it should be
taken "as is" for the basis of a standard for interval arithmetic, I am less
convinced.
Nate