Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!usc!elroy.jpl.nasa.gov!decwrl!world!iecc!compilers-sender
From: preston@ariel.rice.edu (Preston Briggs)
Newsgroups: comp.compilers
Subject: Re: A question about subsumption in register allocators
Keywords: optimize, registers question
Message-ID: <1991Apr12.190307.18265@rice.edu>
Date: 12 Apr 91 19:03:07 GMT
References: <9104081234.AA00418@birk.cs.chalmers.se>
Sender: compilers-sender@iecc.cambridge.ma.us
Reply-To: preston@ariel.rice.edu (Preston Briggs)
Organization: Rice University, Houston
Lines: 26
Approved: compilers@iecc.cambridge.ma.us

Thomas Johnsson <johnsson@cs.chalmers.se> writes:
>Before the actual coloring, subsumption is normally done, to remove
>unnecessary copy operations, and to make the graph to be coloured
>smaller.
>
>Now the question is: What do register allocators out there do if there
>is more than one way of doing the subsumption, and they are mutually
>incompatible?

One of two things:

	Start at the top of the routine and head toward the bottom,
	coalescing as you go.

or

	Find the loop structure (need for spill costs anyway)
	and examine the most deeply nested loops first.

There's very little cost difference between the two.
I don't know about the payoff.

Preston Briggs
-- 
Send compilers articles to compilers@iecc.cambridge.ma.us or
{ima | spdcc | world}!iecc!compilers.  Meta-mail to compilers-request.