Path: utzoo!attcan!uunet!husc6!rutgers!cs.utexas.edu!sm.unisys.com!ism780c!news
From: news@ism780c.isc.com (News system)
Newsgroups: comp.arch
Subject: Re: Quadruple-Precision Floating Point
Message-ID: <20707@ism780c.isc.com>
Date: 21 Dec 88 20:56:32 GMT
References: <8561@alice.UUCP> <400001@uxg.cso.uiuc.edu>
Reply-To: marv@ism780.UUCP (Marvin Rubenstein)
Organization: Interactive Systems Corp., Santa Monica CA
Lines: 17

In article <400001@uxg.cso.uiuc.edu> phil@uxg.cso.uiuc.edu writes:
>I wish some machine had truly VARIABLE precision firmware floating AND FIXED
>point (i.e with fractions) arithmetic.

One machine did provide just this.  It was the IBM/1620.  The precision of
both fixed point (not necessarily integer) and floating point arithmetic was
varaible from 2 decimal digits up to a maximum limited only by the amount of
main memory available. (Yes I said DECMAL digits).  The FORTRAN compiler
provided for integer precision of 4 to 10 digits and floating point precision
of 6 to 20 digits.   The main reason for the relatively small upper bound on
floating precision was the difficulty of providing a math library of
unlimited precision.

The machine was a little slow by todays standards.  An integer add of 5 digits
took 580 microseconds (480 on the model 2).

      Marv Rubinstein