Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site mmintl.UUCP
Path: utzoo!linus!philabs!pwa-b!mmintl!franka
From: franka@mmintl.UUCP (Frank Adams)
Newsgroups: net.arch
Subject: Re: Re: RISC (really on multiplication d
Message-ID: <493@mmintl.UUCP>
Date: Tue, 16-Jul-85 14:57:01 EDT
Article-I.D.: mmintl.493
Posted: Tue Jul 16 14:57:01 1985
Date-Received: Sat, 20-Jul-85 04:28:54 EDT
References: <149@mips.UUCP> <600005@pbear.UUCP>
Reply-To: franka@mmintl.UUCP (Frank Adams)
Organization: Multimate International, E. Hartford, CT
Lines: 17
Summary: Multiply instructions aren't defined right, anyhow



With all the discussion about multiplies, let me express one of my pet
peeves.  Which is that integer multiplies are hardly ever (never, in my
experience) defined in a manner suitable for their most common use.
I would guess, conservatively, that well over 90% of the integer multiplies
that are done generate a result with no more bits than the larger of the
two operands.  That is, one multiplies two words and wants a word result.
But hardware multiplies invariably generate a two word result, leaving the
high-order word to be allowed for and/or disposed of.  This is a problem
equally for compilers and hand-written assembler code.

The same problem occurs for division.

Incidently, re the subscript computation problem, why not a command which
multiplies a word by a one-byte constant?  This would deal cleanly with
most two-dimensional arrays and arrays of structures, yet should be
implementable as a reasonably efficient instruction.