Path: utzoo!yunexus!davecb
From: davecb@yunexus.UUCP (David Collier-Brown)
Newsgroups: comp.arch
Subject: Re: Register usage
Summary: Compilers **can** do better, but if you don't have a good one
Keywords: profile, register allocation
Message-ID: <1963@yunexus.UUCP>
Date: 18 May 89 02:28:10 GMT
Article-I.D.: yunexus.1963
References: <952@aber-cs.UUCP>
Reply-To: davecb@yunexus.UUCP (David Collier-Brown)
Organization: Bimodal: [Interleaf Canada | York U. Computing Services]
Lines: 30

In article <952@aber-cs.UUCP> pcg@cs.aber.ac.uk (Piercarlo Grandi) writes:
[responding to some factual information on register usage]
>I thank you enormously. This quote greatly cheers me up. This is a very good
>support for my position on "register" (where a competent programmer is
>expected to do this, with great simplication of the compiler, because
>compilers cannot possibly know which variables are the most frequently used
>dynamically).

  True if you define "compiler" narrowly (ie, normally).  However, if
you think of an optimization "environment", you can justify feeding the
results of a profile back into the compiler to guide register allocation.
  I would tend to structure this as an optional "pass" (well, pseudo-pass)
for purely practical reasons...

  And if you want to experiment with the idea right now, consider
writing an output processor for a profiler which orders variables by
use in individual functions, and then another utility to rewrite the
source code with register declarations for the X hottest variables
per function iff the call cost/run cost ratio would not be deranged
thereby.  (It is fairly hard (:-)).

--dave (only two more days to monomania!) c-b
--
David Collier-Brown,  | {toronto area...}lethe!dave
72 Abitibi Ave.,      |  Joyce C-B:
Willowdale, Ontario,  |     He's so smart he's dumb.
CANADA. 223-8968      |