Xref: utzoo comp.sys.mac:11249 comp.sys.amiga:13351
Path: utzoo!utgpu!water!watmath!clyde!rutgers!cmcl2!nrl-cmf!ames!oliveb!sun!pepper!cmcmanis
From: cmcmanis%pepper@Sun.COM (Chuck McManis)
Newsgroups: comp.sys.mac,comp.sys.amiga
Subject: Re: Code Optimization
Message-ID: <39879@sun.uucp>
Date: 25 Jan 88 18:08:18 GMT
References: <4776@watdragon.waterloo.edu>
Sender: news@sun.uucp
Reply-To: cmcmanis@sun.UUCP (Chuck McManis)
Distribution: comp
Organization: Sun Microsystems, Mountain View
Lines: 36

In article <4776@watdragon.waterloo.edu> (Paul Larson) writes:
> Currently, single pass compilers are very popular, due to their speed.  This
> speed allows the development cycle to be shortened, sometimes by very
> significant amounts.  However, unless I'm very wrong, the speed of the
> single pass compiler is gained at the expense of code quality.  One 
> immediately obvious solution is a code optimizer, which could be used on a
> compiled program once the critical debuggion is done. ...
>                                                           ...  Thus, if a 
> system is using the more advanced '020, it is using code which was written for
> a more primitive, though similar, processor and thus the code does not take
> proper advantage of the more advanced features of the more modern processors.
> The same agrument applies for the math co-processors, the 68881 and its new
> sibling the 68882, which are rapidly becoming more popular on home computers.

Interesting point Paul, a couple of comments :

On the Amiga the Lattice C compiler optimizes to use code that would be faster
on the '010 - '020 by using instructions, like dbra, that take advantage of the
caching aspects of these chips. Since this does not impose a penalty on the
68000 code it seems to be a good optimization choice. 

And yes, incremental and single pass compilers cannot implement some 
optimization schemes but they are fast at compiling. So they are really good
for prototyping. Thus I think a more reasonable solution would be to have,
in effect, two compilers living in the same executable. One the incremental
compiler, the other a multipass optimizing compiler. Then you offer the user
a choice of which mode to compile the program in. I believe this is the 
approach that Microsoft has taken with Quick-C. Claiming that it is 100%
code compatible with their regular C compiler, so that once you have a 
working program you can recompile with their regular compiler to get optimized,
space efficient code. 


--Chuck McManis
uucp: {anywhere}!sun!cmcmanis   BIX: cmcmanis  ARPAnet: cmcmanis@sun.com
These opinions are my own and no one elses, but you knew that didn't you.