Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!usc!snorkelwacker!mit-eddie!uw-beaver!zephyr.ens.tek.com!tekcrl!tekgvs!jans
From: jans@tekgvs.LABS.TEK.COM (Jan Steinman)
Newsgroups: comp.lang.smalltalk
Subject: Re: Objectworks for Smalltalk-80
Message-ID: <7064@tekgvs.LABS.TEK.COM>
Date: 9 Mar 90 18:37:25 GMT
Organization: Tektronix Inc., Beaverton, Or.
Lines: 22

<khaw@parcplace.com (Mike Khaw)>

<Just to clarify, when you "accept" a new method in Smalltalk-80 the compiler 
compiles it into byte codes.  When that method is invoked, the byte code 
representation is dynamically translated to machine code and kept in a "native 
method" cache. The byte code representation saves space, while dynamic 
translation is more efficient than interpreting the byte codes.>

Assuming you are going to use the method again before the cache flushes, native 
code via "dynamic translation" is faster.  However, DT has it's own overhead, 
which must be amortized over a number of executions just to break even.  DT 
works great on things that loop a lot -- like benchmarks.

Interpreting byte codes is not very expensive if the interpreter is tightly 
coded in assembler.  PPS has chosen a C implementation for portability.  On 
typical, UI intensive Smalltalk code, I find an efficient assembly coded 
interpreter to be faster than a C interpreter with DT.

							   Jan Steinman - N7JDB
					Tektronix Electronic Systems Laboratory
					Box 500, MS 50-370, Beaverton, OR 97077
						(w)503/627-5881 (h)503/657-7703