Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!snorkelwacker!spdcc!esegue!compilers-sender
From: larus@spool.cs.wisc.edu
Newsgroups: comp.compilers
Subject: Re: Unsafe Optimizations (WAS: Compiler Design in C How about it?)
Keywords: optimize
Message-ID: <1990Jun4.212226.18389@esegue.segue.boston.ma.us>
Date: 4 Jun 90 21:22:26 GMT
References: <1990Jun4.044255.14857@esegue.segue.boston.ma.us> <1990Jun1.194941.5781@esegue.segue.boston.ma.us>
Sender: compilers-sender@esegue.segue.boston.ma.us
Reply-To: larus@spool.cs.wisc.edu
Organization: University of Wisconsin--Madison
Lines: 18
Approved: compilers@esegue.segue.boston.ma.us
Original-sender: news@spool.cs.wisc.edu


Just for the sake of argument, let me disagree with the prevailing sentiment
that compilers should only provide "safe" optimizations. Keppel has already
raised the argument that some programs are too slow to run without potentially
dangerous optimizations.  This is an interesting argument that brings up the
real point: programming languages are not designed for parallel computation.
Compilers for machines with parallelism adhere to a standard that is too high:
sequential equivalence.  Some programmers are willing to trade off the
semantics of the language (the effect of "bad" optimizations) for faster
programs.  By arguing that compilers should only perform conservative
optimization, you are claiming that the sequential semantics of FORTRAN (fill
in your favorite or least favorite language) are suitable for parallel
execution.  Think carefully before you argue this position.


/Jim
-- 
Send compilers articles to compilers@esegue.segue.boston.ma.us
{spdcc | ima | lotus}!esegue.  Meta-mail to compilers-request@esegue.
Please send responses to the author of the message, not the poster.