From mboxrd@z Thu Jan  1 00:00:00 1970
X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on polar.synack.me
X-Spam-Level: *
X-Spam-Status: No, score=1.3 required=5.0 tests=BAYES_00,INVALID_MSGID,
	MSGID_RANDY autolearn=no autolearn_force=no version=3.4.4
X-Google-Language: ENGLISH,ASCII-7-bit
X-Google-Thread: 103376,ef33c33c4f98bde1
X-Google-Attributes: gid103376,public
From: Robert Dewar <robert_dewar@my-deja.com>
Subject: Re: Compiler for Z80/6510
Date: 1999/11/27
Message-ID: <81nmu0$e2n$1@nnrp1.deja.com>#1/1
X-Deja-AN: 553509734
References: <slrn83nahf.m1.lutz@taranis.iks-jena.de>
 <383c6fed.458467@news.fiam.net> <slrn83ptfb.kn.lutz@taranis.iks-jena.de>
 <81k67s$47l$1@nnrp1.deja.com> <383DC86C.19A6F176@australia.boeing.com>
 <81m4m4$ci0$1@nnrp1.deja.com> <wcc3dtsvqy2.fsf@world.std.com>
X-Http-Proxy: 1.0 x43.deja.com:80 (Squid/1.1.22) for client 205.232.38.14
Organization: Deja.com - Before you buy.
X-Article-Creation-Date: Sat Nov 27 04:33:03 1999 GMT
X-MyDeja-Info: XMYDJUIDrobert_dewar
Newsgroups: comp.lang.ada
X-Http-User-Agent: Mozilla/4.04 [en] (OS/2; I)
Date: 1999-11-27T00:00:00+00:00
List-Id: <comp.lang.ada>

In article <wcc3dtsvqy2.fsf@world.std.com>,
  Robert A Duff <bobduff@world.std.com> wrote:
> Do you completely redo the hand-crafted register allocation
> (which indeed is very efficient)?  Or do you minimize the
amount of code
> that has to be changed, thereby reducing the probability of
bugs?
> In many cases, it's irresponsible to do the former.


Well the answer is that if you are aiming at optimal code,
you certainly redo the register allocation.

I remember one incident when Ken Belcher and I were writing
SPITBOL/370 which is entirely in BAL.

There was a bug that needed fixing in a routine that was
maybe 100 lines long, and we both kept putting it off.

Finally I came in one day, and said

"I fixed that bug ....

 but there is an extra LR in the middle of the routine"

Ken answered immediately "Oh well if you are willing to leave
in junk code like that the fix is trivial", and we both new
without discussion EXACTLY the issue, namely that the register
allocation had to be completely redone for the entire routine
to avoid this one register-register move in the middle of it.

Eventually it got fixed properly :-)


But this is not to suggest for a moment that ANYTHING should be
written at this level these days, we can afford to waste some
machine time to avoid the maintenance nightmares that Bob quite
rightly alludes to!

Still it was amazing what could be achieved (the SPITBOL lexer
and parser used the TRT instruction extensively, and SPITBOL
compiles at millions of lines a minute on a modern mainframe).


Sent via Deja.com http://www.deja.com/
Before you buy.