Path: g2news2.google.com!news3.google.com!feeder.news-service.com!85.214.198.2.MISMATCH!eternal-september.org!feeder.eternal-september.org!.POSTED!not-for-mail
From: D Herring <dherr...@at.tentpost.dot.com>
Newsgroups: comp.lang.lisp
Subject: Re: CPS and optimization
Date: Wed, 10 Nov 2010 01:34:01 -0500
Organization: A noiseless patient Spider
Lines: 19
Message-ID: <ibdecq$gck$1@news.eternal-september.org>
References: <ibd50j$gsr$1@news.eternal-september.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Injection-Date: Wed, 10 Nov 2010 06:34:02 +0000 (UTC)
Injection-Info: mx02.eternal-september.org; posting-host="ydozjAGyeA7MLfD76BLrUw";
	logging-data="16788"; mail-complaints-to="ab...@eternal-september.org";	posting-account="U2FsdGVkX19XY3pNRUKNXmVvgur0P3tO"
User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.2.9) Gecko/20100916 Thunderbird/3.1.4
In-Reply-To: <ibd50j$gsr$1@news.eternal-september.org>
Cancel-Lock: sha1:WgoXyzhc4sHX4ByJZ1dnaRTKVqw=

On 11/09/2010 10:53 PM, Peter Keller wrote:

> I know of the correspondance between CPS and SSA forms, is that what I
> would apply here? I'd convert the CPS form to SSA basic blocks, perform
> the optimization operations upon it, then convert it back to CPS?

Rather each SSA optimization should have an equivalent optimization 
for "well behaved" CPS forms.  The well-behaved restriction comes from 
CPS allowing some things that are not possible in SSA.


[OT] If you do convert to SSA, I'd just throw the code through LLVM. 
It seems to me that low-level optimizations (e.g. local scheduling, 
opcode selection, and register allocation) are fairly well solved 
whereas higher-level optimizations (e.g. profiling-driven inlining, 
fast-path generation, and data-structure selection) need more 
interaction with the programmer and provide interesting new opportunities.

- Daniel