Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!rutgers!ames!sdcsvax!maiden
From: maiden@sdcsvax.UCSD.EDU (VLSI Layout Project)
Newsgroups: comp.sys.mac
Subject: code optimizers
Message-ID: <3924@sdcsvax.UCSD.EDU>
Date: Mon, 21-Sep-87 13:14:36 EDT
Article-I.D.: sdcsvax.3924
Posted: Mon Sep 21 13:14:36 1987
Date-Received: Tue, 22-Sep-87 04:56:43 EDT
Organization: University of California, San Diego
Lines: 45



To the Macintosh Development Community:

Is there a significant interest market for an external optimizer for
the Macintosh/Macintosh II?  Recently some CS associates and I have
discussed the possibility of applying that 'ol computer science
knowledge toward developing such a product.

What do you, the development community, think?  Are existing compilers
sufficiently endowed with the ability to produce fast code?  Are those
developers who have bemoaned the lack of code quality on the Macintosh
comparable to that on other computers (including the PC) willing to
put their money where their mouth is?  IS THERE REALLY A MARKET FOR
AN OPTIMIZER - or are developers willing to exploit alternative
routes (hand-coded assembly by an expert)?

Also: are there such products already being developed?  There are
other things far easier (and more fun!) to do than making an 
optimizer; if someone is doing it already, we will willingly step 
aside and do something else.

And a final question: what are the specifications of an ideal code
optimizer?  What are the code size/speed tradeoffs that developers
out in the Macintosh community desire?  Obviously the optimizer
would be modular, allowing selective enabling/disabling of the
various levels of optimization, but is there a need for optimization
of ROM calls or cache use or coprocessor parallelism?

Presently we would envision the system to run under MPW - are there
objections to this? - and operate on assembler output to maximize
its functionality among several compilers.  We would consider
implementing loop optimization, induction variable and global common 
subexpression elimination, loop unrolling and merging, constant
folding, copy propagation, code hoisting, procedure call optimization,
etc.  It may be desirable to allow directives to be passed to the 
optimizer for greater control and implementation of other types of
optimization, including code substitution and algorithm optimization.

Your suggestions welcome.  E-mail is preferred, however I often
scan this board.

Edward
------------------------------------------------------------------------
UUCP: {seismo|decwrl}!sdcsvax!maiden     ARPA: maiden@sdcsvax.ucsd.edu