Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!mit-eddie!uw-beaver!tektronix!oresoft!rick
From: rick@oresoft.UUCP (Rick Lahrson)
Newsgroups: comp.sys.amiga,comp.lang.misc
Subject: Re: Compilers?
Message-ID: <71@oresoft.UUCP>
Date: Wed, 31-Dec-69 18:59:59 EDT
Article-I.D.: oresoft.71
Posted: Wed Dec 31 18:59:59 1969
Date-Received: Wed, 16-Sep-87 02:23:29 EDT
References: <1708@amiga.amiga.UUCP> <581@sugar.UUCP> <935@unicus.UUCP> <66@oresoft.UUCP> <5042@jade.BERKELEY.EDU> <70@oresoft.UUCP> <2122@sol.ARPA>
Reply-To: rick@oresoft.UUCP (Rick Lahrson)
Organization: Oregon Software, Portland OR
Lines: 40
Keywords: compiler assembler dwim mating
Summary: Just learning to talk "net" 8-)
Xref: mnetor comp.sys.amiga:8356 comp.lang.misc:659

In article <2122@sol.ARPA> crowl@cs.rochester.edu (Lawrence Crowl) writes:
>In article <70@oresoft.UUCP> rick@oresoft.UUCP (Rick Lahrson) writes:
>>I say the compiler should model the target hardware, and generate
>>code for that.
>
>Wait a second!  We have two models, the language model and the machine model.
>The compiler itself does not model anything, it translates between two models.
>I submit the language should not model any SPECIFIC machine.  (Note that most
>algorithmic languages assume a rather generic von Neuman architecture.)
>
>The compiler should implement the language's model on the machine's model as
>best as possible given the development resources.  As long as the language
>model is faithfully implemented, how close the resulting code is to the machine
>model is a performance issue.  (All bets are off if you are mixing languages.)

OK, I'm starting to catch on.  I'm a relative neophyte at posting to the net,
with its wide ranging readership (I'm used to discussing such topics with just
my fellow compiler writers, and there's a large body of things understood as
background, so my communications have come to rely on that -- I'll be more
rigorous in the future).

Yes, compilers deal with two models: the language and the target machine (or
pseudo-machine -- in this case the 68K, a real machine).  When I said the
compiler should model the target hardware, I meant that it should use its
"mental model" of the target to determine how to generate code.  I didn't
suggest that a LANGUAGE should have anything to do with specific hardware,
just the compiler (specifically, just the "back end" (code generator) of the
compiler).  And yes, it is exactly a performance issue.  That's what offends
me about BPTR's -- they waste time and space.  I don't know BCPL, and maybe
that voids my right to an opinion about BPTR's, but certain compilers I DO
know about can actually translate "add one word" into "add four bytes"! 8-)
Seems to me BCPL should have been able to do that, too.

Sure, I'm willing to let this topic die. 8-)

-- 

Rick Lahrson  ...tektronix!oresoft!rick

Disclaimer:  If I ever speak for anyone but me, I'll warn you in advance.