Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!mit-eddie!uw-beaver!cornell!rochester!PT.CS.CMU.EDU!sei!sei.cmu.edu!firth
From: firth@sei.cmu.edu (Robert Firth)
Newsgroups: comp.lang.c
Subject: Re: 16x16 = 32
Message-ID: <2884@aw.sei.cmu.edu>
Date: Fri, 16-Oct-87 09:24:23 EDT
Article-I.D.: aw.2884
Posted: Fri Oct 16 09:24:23 1987
Date-Received: Sat, 17-Oct-87 20:09:14 EDT
References: <143@kesmai.COM>
Sender: netnews@sei.cmu.edu
Reply-To: firth@bd.sei.cmu.edu.UUCP (Robert Firth)
Organization: Carnegie-Mellon University, SEI, Pgh, Pa
Lines: 33

In article <143@kesmai.COM> dca@kesmai.COM (David C. Albrecht) writes:

>In virtually every machine I know of the multiplies produce a
>double width result (PDP-11, VAX, 68000, 8?86, ad nauseum)

(1) On the PDP-11, the MUL instruction produces a single-length
    result if the destination is an odd-numbered register, and
    a double-length result if it's an even-numbered register

(2) On the Vax-11, the MULx instructions all produce single-length
    results.  A double-length result requires either first widening
    byte or word, or using EMUL for longword.

(3) On the MC68020, the result of MULS is either single-length or
    double-length, depending on the Sz field of the instruction;
    the lower members of the range do not have a hardware 32-bit
    multiply.

And with that piece of gratuitous misinformation out of the way, to the
point.  The task of a compiler is not to provide a more readable form
of the machine's instruction set.  It is to implement the language as
defined in the standard.  If the standard says that int*int => int, then
that's what you implement, and anything else is plain WRONG.  If the
standard says you are allowed to keep intermediate results to higher
range or precision, then you may do so; but you would be well advised
to understand the consequences - in particular, that seemingly harmless
changes to a working program, like taking an expression out of line, or
doing common subexpression elimination by hand, may break a working
program.

When in doubt, implement the canonical semantics.  At the very least, it
is a courtesy that will be appreciated by the user who has taken the
trouble to read the reference manual.