[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Float Pt Precision



"Michael J. Donahue" <donahue@saaz.cam.nist.gov> writes:

 > The difference in the floating point calculations on the x86 vs the
 > Alpha trace back to the hardware implementations.  The extra
 > precision in the x86 makes life much easier---it narrows the window
 > of numerical instability in almost every situation.

Can you substantiate this?  I've found x86 arithmetic to be
substantially harder to work with because the excess precision shows
up unpredicably.  After setting two different variables via the same
computation it's possible for the two variables to be not equal.  It
depends on exactly what goes into registers when.  This problem
doesn't exist on the alphas.

-- 
Harvey J. Stein
Bloomberg LP
hjstein@bfr.co.il



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index] []