Re: MidRad -- and Two Different Application Domains (TDAD).
On 2009-09-21 06:57:47 -0400, Michel Hack wrote:
> Vincent Lef?vre brought up the example
> > ... And what about expressions like A + B + C? Do you have
> > a canonical definition that would satisfy all the users?
> > For instance, if A = x and C = -x (x as a variable), would
> > you allow an implementation to optimize this expression to B?
>
> This is a fine example as to why I believe it is important to state
> explicitly (e.g. to the compiler, via pragma) which of the two
> application domains applies: if an interval stands for a single
> but uncertain numerical value, the simplification is legitimate,
> but if the intervals represent domains, it would depend on extra
> information, and in general would not be legitimate.
Yes, nice point.
I was thinking of the former case, where the simplification is
legitimate. Even in this case, if the goal is to obtain reproducible
results, the simplification must not be done, and this can be a
performance problem.
--
Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arénaire project (LIP, ENS-Lyon)