Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!husc6!spdcc!ima!compilers-sender From: pardo@june.cs.washington.edu (David Keppel) Newsgroups: comp.compilers Subject: Re: a practical UNCOL Summary: Impedance matching Keywords: compiler ir rtl dag code generation Message-ID: <3934@ima.ima.isc.com> Date: 13 May 89 12:57:26 GMT References: <3892@ima.ima.isc.com> Sender: compilers-sender@ima.ima.isc.com Reply-To: David Keppel Organization: U of Washington, Computer Science, Seattle Lines: 24 Approved: compilers@ima.UUCP Return-Path: Ralph Johnson writes: >[Fraser/Davidson RTL IR] >[moderator: isn't that what gcc uses (successfully)?] My compilers prof has pointed out to me that matching for optimizations and instruction selection in an RTL (Fraser/Davidson, IBM PL.801) is a special case of matching in a DAG (directed acyclic graph) -structured ir (Graham/Glanville/Henry). One important distinction is that all matching in an RTL is done at the level of things that usually look like RISC machine instructions. Thus the code quality is better when there is a good match between the instructions in the machine-independent RTL and the instructions in the physical hardware. ;-D on ( The code degenerator ) Pardo -- pardo@cs.washington.edu {rutgers,cornell,ucsd,ubc-cs,tektronix}!uw-beaver!june!pardo -- Send compilers articles to compilers@ima.isc.com or, perhaps, Levine@YALE.EDU Plausible paths are { decvax | harvard | yale | bbn}!ima Please send responses to the originator of the message -- I cannot forward mail accidentally sent back to compilers. Meta-mail to ima!compilers-request