Path: utzoo!attcan!uunet!husc6!mailrus!cornell!uw-beaver!teknowledge-vaxc!sri-unix!quintus!ok
From: ok@quintus.uucp (Richard A. O'Keefe)
Newsgroups: comp.sys.apollo
Subject: Re: Fortran 9.95 problem
Message-ID: <745@quintus.UUCP>
Date: 24 Nov 88 09:43:28 GMT
References: <8811231951.AA29470@umix.cc.umich.edu>
Sender: news@quintus.UUCP
Reply-To: ok@quintus.UUCP (Richard A. O'Keefe)
Organization: Quintus Computer Systems, Inc.
Lines: 13

In article <8811231951.AA29470@umix.cc.umich.edu> derstad@CIM-VAX.HONEYWELL.COM ("DAVE ERSTAD") writes:
>    A = something
>    B = DSQRT(A*A + C*C)
>    D = A + B - C
>    IF D = 0 jump over some code

Just a word in defence of Apollo:  I ran into the same problem with a
PR1ME 400 years ago.  (Single-precision calculations were done in a
double-precision register, then truncated, except optimisation skipped
the truncation.)  A similar problem also exists on Suns, which is why
they have the "-fstore" option.  I think Apollo have a similar option.

That isn't a terribly brilliant way of computing B, but that's another story.