Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!usc!zaphod.mps.ohio-state.edu!sdd.hp.com!decwrl!bacchus.pa.dec.com!decvax.dec.com!ima!esegue!compilers-sender
From: david@glance.ch (David Mosberger)
Newsgroups: comp.compilers
Subject: How to read/print fp numbers accurately
Keywords: design, Fortran, question
Message-ID: <9008211656.AA10898@chx400.switch.ch>
Date: 22 Aug 90 05:22:37 GMT
Sender: compilers-sender@esegue.segue.boston.ma.us
Reply-To: david@glance.ch (David Mosberger)
Organization: Compilers Central
Lines: 23
Approved: compilers@esegue.segue.boston.ma.us

Recently, two articles appeared in a SIGPLAN paper concerning the optimal
conversion of floating point numbers represented in a decimal scientific
notation to binary floating point representation and back. ``Optimal'' is
meant in the sense of ``best approximation to the true binary/decimal
value''. The presented algorithms were very elaborate. However, they
require either multi-precision integers or extended precision
floating-point operations or both.

I would like to know, what the best is one can get using single precision
floating point only? I.e., if you want to convert to/from single precision
floating point numbers, the algorithm should only use single precision
floating point operations (of course, integers with a ``usual'' size may
be used as well).  Is there an optimality criterion for such an algorithm?

David Mosberger						Glance Ltd.
Software Engineer					Gewerbestrasse 4
david@glance.ch						8162 Steinmaur
UUCP: {...}!{uunet,mcsun}!elava!david			Switzerland
X.400: S=david;O=glance;P=switch;A=arCom;C=ch
BITNET: david@glance.ch or david at glance.ch
-- 
Send compilers articles to compilers@esegue.segue.boston.ma.us
{ima | spdcc | world}!esegue.  Meta-mail to compilers-request@esegue.