Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uwm.edu!zaphod.mps.ohio-state.edu!rpi!bu.edu!mirror!frog!cpoint!crackers!m2c!umvlsi!dime!smectos!eli
From: eli@smectos.gang.umass.edu (Eli Brandt)
Newsgroups: comp.lang.pascal
Subject: Re: TP: Exponentiation?
Message-ID: <17410@dime.cs.umass.edu>
Date: 25 Jul 90 22:05:00 GMT
References: <950036@hpclapd.HP.COM> <2041@bimacs.BITNET>
Sender: news@dime.cs.umass.edu
Reply-To: eli@smectos.CS.UMASS.EDU (Eli Brandt)
Organization: University of Massachusetts, Amherst
Lines: 17

In article <2041@bimacs.BITNET> ehrlich@bimacs.biu.ac.il.UUCP (Gideon Ehrlich) writes:
>
>Very easy. Either use    x**y = exp(y*ln(x))
>           or write your own function ,using Tailor sequence.

The power function is fine, but you should really never use a Taylor sequence for
*anything* other than maybe prototyping.  If you write your own power function, 
you should consider these modifications:
   1) Check for negatives and 0^0, obviously.
   2) A significant fraction of calls will probably involve integer powers.  The
      way to handle this is to use a simple for loop for low powers and a repeated
      squaring loop for higher powers.  The crossover point depends on the machine
      and compiler but is usually around 5 to 10.
   3) If you have a power y which is i+r, where i is floor(y), you can use your
      integer-power routine and multiply the result by the output of a power
      routine optimized for its fixed domain.  I didn't get any significant
      speedup with this, though.