Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

754-2008 extendable precisions



Vincent LefÛvre wrote:
> For instance, how would you compute sin(2^emax)?

That's a no-brainer in IA:  return [-1,+1].

Indeed, unless an input argument is strictly a point interval,
that's the required result as soon as an ulp exceeds Pi, at
least with InfSup representations.  (MidRad enthusiasts can
point out with glee that they may be able, with enough hard
work, to give better results here!)

Michel.

P.S.  I changed the "Subject:" line (again).  Motion M0002
      has been getting awfully many YES votes lately...
Sent: 2009-03-24 12:52:43 UTC