[antlr-interest] some quick codegen optimization thoughts

[Terence Parr]

Howdy,

So I've spent some time implementing switches vs if-then-else chains 
(in Java and bytecodes (yes, that was a pain)).  Further, I have worked 
on factoring out input.LA(1) calls.  The code looks pretty clean I 
think.  I've been using the java.g grammar as a benchmark.  The 
decision making is optimal in that it references input.LA(k) at most 
once per k value need to predict alternatives (both with if-then-elses 
and switches; both in Java and in bytecodes).

The summary is this.  Generating switches generates slightly bigger 
code .class files.  It seems to gain only about 10% in CPU time, which 
ain't much given all the effort I just put in. ;)  I'm guessing C and 
other compilers will be able to take better advantage of my efforts. ;) 
  To give you a benchmark against 2.x ANTLR doing the same grammar (but 
doing tree construction), ANTLR's generated parser is only about 2x as 
fast, but that ain't too bad to start.  I have one really major 
optimization left to do (later) that should make it much better.

I'm rapidly closing in on a complete piece of software for an 
early-release.  Then I'll need some more sample grammars so people can 
actually try it out.  Still I'm thinking late May timeframe.

Ter
--
CS Professor & Grad Director, University of San Francisco
Creator, ANTLR Parser Generator, http://www.antlr.org
Cofounder, http://www.jguru.com