numeric Keith Bierman on signaling NaNs

[David G Hough at validlab.com]

This bounced when posted because Sun's email addressing convention changed,
so I'm reposting it for Keith:

 *****************************************************************

Date: Fri, 23 Aug 2002 15:42:55 -0700
From: keith bierman <Keith.BiermanaSun.COM>
Subject: Re: numeric question about usage of signaling NaNs


"David G Hough at validlab.com" wrote:
> 
> At the 754R meeting yesterday, a question arose: is anybody actually
> using 754's signaling NaNs?   

Many years ago, in a company far, far away, we used the signaling
NaN as a way to get the system to halt. I don't recall why that was
the best way to do it, but it seemed to make sense at the time.


>The only actual usage those present at the
> meeting were familiar with was using signaling NaNs to initialize storage
>

I'm pretty sure that some of those debugging techniques leverage the
signaling NaN .. so if these were taken away a new method would be
required.
 
>> A number of aspects of 754, such as conversion between binary and
> decimal character strings, are complicated by the need to support signaling
> NaNs, to the extent that 754R contemplates removing the requirement that
> implementations provide signaling NaN patterns.

Certainly a very simplfied signaling NaN (just one, rather than a
family of patterns) would suffice for any application I can recall.
I'm not sure that those applications would care if base conversion
ignored the potentially signalling stuff. If it really showed up in
base conversion (which actually got written out) first and then
halted, I'd actually prefer it ... but that might be a little
perverse.

-- 
Keith H. Bierman    keith.biermanaSun.COM| 650-352-4432 voice+fax
Sun Microsystems Laboratories            | sun internal 68207
15 Network Circle  UMPK 15-224           | 
Menlo Park, California  94025 	         | kbiermanaacm.org
            <speaking for myself, not Sun*> Copyright 2002