home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!charon.amdahl.com!pacbell.com!sgiblab!darwin.sura.net!Sirius.dfn.de!hpux.rz.uni-jena.de!pfk
- From: pfk@rz.uni-jena.de (Frank Klemm)
- Subject: Re: integer distance or sqrt
- Message-ID: <1992Nov17.171815.7039@rz.uni-jena.de>
- Sender: pfk@rz.uni-jena.de (Frank Klemm)
- Reply-To: pfk@rz.uni-jena.de (Frank Klemm)
- Organization: Friedrich-Schiller-University Jena, Germany
- References: <55F0TB1w165w@tsoft.sf-bay.org> <CLIFFC.92Nov15184340@miranda.rice.edu>
- Date: Tue, 17 Nov 1992 17:18:15 GMT
- Lines: 24
-
-
- In article <CLIFFC.92Nov15184340@miranda.rice.edu>, cliffc@rice.edu (Cliff Click) writes:
- |> bbs.dennis@tsoft.sf-bay.org (Dennis Yelle) writes:
- |> > because of the interest in Herman Rubin's optimization problems
- |> > some people here might be interested in this.
- |> > It is a real problem. That is, I have a program that spends quite a
- |> > lot of time here, and I would like to change this if I could.
- |>
- |> > I first considered asking for a integer sqrt, but maybe if the context is
- |> > known a better answer can be given. This program is normally run on a
- |> > machine with no floating point hardware so I would like to get rid of
- |> > floats entirely.
- |>
- |> I sent him an integer square root code using Newton's Method. I've done
- |> this problem before and I'm sure there are better methods out there. Does
- |> somebody want to post/email something better than Newton's?
- |>
- I have programmed several methods on a i386. Newton is the fastest if the mul instruction
- is so slow as on a Intel. If multiplication is fast and division is slow, bit trying
- and a newton cycle is an optimum. So I need the time of the div and mul and if the instruction
- exist, of a bit search instruction to evalutate this problem. Is an error of a LSB critical (sqrt(24)=5) is another
- question.
-
- Frank
-