Path: utzoo!attcan!utgpu!news-server.csri.toronto.edu!clyde.concordia.ca!uunet!snorkelwacker!spdcc!ima!esegue!compilers-sender
From: larus@primost.cs.wisc.edu (James Larus)
Newsgroups: comp.compilers
Subject: Re: Intermediate Representation
Keywords: code, optimize
Message-ID: <1990Aug12.134822.10054@esegue.segue.boston.ma.us>
Date: 12 Aug 90 13:48:22 GMT
References: <1990Aug07.153407.8877@esegue.segue.boston.ma.us>
Sender: compilers-sender@esegue.segue.boston.ma.us
Reply-To: larus@primost.cs.wisc.edu (James Larus)
Organization: U of Wisconsin CS Dept
Lines: 19
Approved: compilers@esegue.segue.boston.ma.us

In my opinion (a phrase sadly missing from other replies to this message),
both RTL and AST/DAG representations are equivalent for a compiler designed
and written for a single machine.  I base this belief on having written or
worked on compilers that use both types of IR. Both forms represent the
program in a form that is easy to traverse and manipulate.  However, I don't
believe that the equivalence holds for retargetable compilers.  RTL appears
to be a much better IR for retargetable code generators.  The
Davidson-Fraser approach produces fast, practical code generators while the
retargetable CG for ASTs are still mainly university projects.

A more interesting question is: how would you demonstrate that one approach
is superior to the other one?  If computer science was a science, we could
just lock up a sample of programmers in a room and have them write a
compiler of a randomly-choosen type and compare the final products.

/Jim
-- 
Send compilers articles to compilers@esegue.segue.boston.ma.us
{spdcc | ima | lotus| world}!esegue.  Meta-mail to compilers-request@esegue.