🚀 go-pugleaf

In article <dbell-1504961320040001@wholder2.cts.com> dbell@shvn.com (Doug Bell) writes:

>In article <goochb.339.00099333@rwi.com>, goochb@rwi.com (William D.
>Gooch) wrote:

>> In article <3170B180.51A8@concentric.net> "Alan L. Lovejoy"
><alovejoy@concentric.net> writes:
>>
>> >Another point to consider: Smalltalk (and LISP) would probably benefit 
from 
>> >a Java processor as much as would Java itself.  At the "bytecode" 
level,
>> >syntax differences and static typing are no longer relevant.>> 

>> Excellent point.  Anyone know how many bits are in a Java
>> machine word?

>Er...A byte is usually 8-bits.  That's what a bytecode is in this case at 
least.

My reference to "Java machine" was about hardware, not the 
virtual machine.  Bytecoding has little to do with this - issues
such as handling 32-bit and 64-bit numeric values efficiently are
more relevant.

Having thought about what Alan said a bit more, I now think that
Java hardware probably won't do a whole lot for Lisp or Smalltalk,
because Java is statically typed.  The same reasoning indicates,
I believe, that the speed improvements to be gained from special-
purpose Java hardware are not likely to be as significant as those
that have been seen for Lisp machines.  This will most likely make it 
harder for Java hardware to compete with conventional hardware 
than it has been for Lisp machines.

Lisp machines, for example, have 36 or more bits/word, to accomodate 
data type and other tag bits along with a full 32-bit numeric value.  
Conventional hardware with 32-bit words obviously can't handle 
tags and still have 32 bits left over for numeric data.  This means that 
Lisp on conventional hardware frequently uses two words where a 
statically typed language only needs one word.

Java hardware would not need data type tags because of static 
typing.  Thus, Java machine words would not be long enough to 
work very well for Lisp (probably the same is true for Smalltalk).
Lisp machines have also made use of parallel type-checking hard-
ware, which again is not needed for a statically typed language.

There are other tag bits on Lisp machine words which contain non-
numeric data.  Pointers use additional tags for "cdr-coding," which
is specifically to support compactness and speed in the handling of 
list structures, which are fundamental in Lisp.  This further limits the
address space, because it uses up some of the remaining 32 bits, 
and this has led to the most recent Lisp hardware architectures 
using a word size greater than 36 bits.

One area where Java hardware may incorporate some enhancements 
that have proved valuable with Lisp machines is garbage collection.

~~~~~~~~~~~~~~~~~~~~~~
William D. Gooch
RothWell International
goochb@rwi.com
Texas liaison for the International Programmers Guild
For information on IPG, see http://www.ipgnet.com/ipghome.htm
~~~~~~~~~~~~~~~~~~~~~~