Re: Motion P1788.1/M004.01
On 2016-05-18 07:24:32 -0300, Walter Mascarenhas wrote:
> ok, I agree with Vincent, and I also find some claims about Unums to
> be a bit magical.
For Unums 2.0, there are also claims that are obviously incorrect. On
http://www.johngustafson.net/presentations/Unums2.0slides-withNotes.pdf
slide 44, Gustafson gives the example:
Let x = [2,4]. Repeat several times x <- x - x; Print x.
He notices that intervals get larger and larger with intervals,
while with his system SORNs (which basically corresponds to a
union of elementary intervals), this converges to 0, so that he
claims that his system is better. With the first iteration, he
gets the interval (-1,1) in his system. However, what he seems
to be unaware of is that if x = [2,4] and y = [2,4], this would
lead to the same operation (at the arithmetic level, the notion
of dependencies between variables is absent), so that he would
also get (-1,1), which is incorrect if the real value of x is
2 and the real value of y is 4.
--
Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <https://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <https://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)