Re: Accuracy of 754-compliant recommended functions
On 2013-07-01 12:44:28 -0400, Michel Hack wrote:
> It is only RELEVANT for DFP, because for BFP the quantum of any given
> representable value is fixed. The quantum has nothing to do with the
> accuracy of a result. The preferred quantum only applies to exact DFP
> results (e.g. sin(0)), because that is the only situation where there
> is a choice of distinct representations for the same value (e.g. 0,
> which has hundreds of distinct representations in DFP).
The "only situation" above refers to "DFP", not to "exact": even
if you get an inexact (rounded) value, this value can have several
distinct representations (but the "preferred" rule is not used in
such a case: the representation must be normalized).
--
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 / AriC project (LIP, ENS-Lyon)