Path: utzoo!attcan!uunet!lll-winken!lll-lcc!ames!mailrus!husc6!ogccse!blake!uw-beaver!uw-june!rik
From: rik@june.cs.washington.edu (Rik Littlefield)
Newsgroups: comp.arch
Subject: Re: Content Addressible Memories
Message-ID: <6746@june.cs.washington.edu>
Date: 16 Dec 88 06:15:29 GMT
References: <00051@meph.UUCP>
Organization: U of Washington, Computer Science, Seattle
Lines: 21

In article <00051@meph.UUCP>, gsarff@meph.UUCP (Gary Sarff) writes:
> 
> The languages SNOBOL4,SPITBOL ... and
> Icon all have associative tables as first class data structures, i.e. they
> can be passed as arguments to functions, returned as results, assigned, 
> compared, an entire table can be an element of a larger aggregate type such
> as a field in a record, an array element etc.  CAM's sound ideal for this
> since these languages pass around descriptors for these tables rather than
> the tables themselves and the elements in the tables are also descriptors
> that would fit into CAM memory locations.
> 
Yes, these tables are very slick.  But when I use them, the "index" values
usually end up being strings of substantially more than 32 or 64 bits,
and the tables very often have more than 256 entries, sometimes several 
thousand.  Furthermore, these languages don't have declarations that would 
allow a compiler to determine when a fixed-size CAM would be big enough,
so there would be execution time overhead in deciding what implementation
to use (pure software or CAM-assisted).  I remain unconvinced that the 
current and foreseeable CAMs would provide enough improvement in this
context to justify their cost.

--Rik