Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site opus.UUCP
Path: utzoo!linus!vaxine!wjh12!harvard!seismo!hao!cires!nbires!opus!rcd
From: rcd@opus.UUCP (Dick Dunn)
Newsgroups: net.lang
Subject: Re: FORTRAN vs. C (and number crunchers)
Message-ID: <593@opus.UUCP>
Date: Tue, 3-Jul-84 19:30:23 EDT
Article-I.D.: opus.593
Posted: Tue Jul  3 19:30:23 1984
Date-Received: Tue, 10-Jul-84 04:45:04 EDT
References: <174@callan.UUCP> <586@opus.UUCP>
Organization: NBI, Boulder
Lines: 28

Please ignore the article 586@opus (same title).  Here's the un-messed-up
version (assuming I can get it put together and get out of the editor
without any more fumbles.

>All of this discussion of the merits of Fortran vs. C seems to be missing one
>of the most important points of all:  ACCURACY.
>
>My point is not that C is inherently inaccurate...
>...I would count on the
>people at Cray to do the best job possible;  after all, it's their bread and
>butter...

Perhaps you haven't seen some of the bizarre phenomena that have come from
Mr. Cray.  The Cray I has no divide instruction; instead it has a
reciprocal approximation (emphasis on "approximation").  Try that one for
numerical analysis.  Or how about his earlier designs, the 6600 and family?
The rounded floating add/subtract round the operands rather than the
result.  The rounded divide is even more exotic; in effect the result is
rounded by either 1/3 or 2/3, depending on whether a final shift is
required to normalize the result.  I haven't had any opportunity to look at
the software since old CDC 6x00 days (late 60's for me), but at that time
the software matched the hardware in numerical soundness.  In other words,
don't believe what the salesmen tell you.  Don't even believe what you
might think ought to be true.  Check it out carefully (and cynically).

-- 
Dick Dunn	{hao,ucbvax,allegra}!nbires!rcd		(303)444-5710 x3086
	...Never attribute to malice what can be explained by stupidity.