Re: [oc] Beyond Transmeta...

[Jecel Assumpcao Jr <jecel@merlintec.com>]

Leyland,

As Jon Beniston mentioned, it is certainly worth studying the old
connection machine designs:

   http://www.jps.net/ethelen/comp-hist/connection-machine.html

Many years ago, I created a few chip designs for executing cellular
automata and some of them were rather close to what you want. I bet
many other people did the same and there might be some infomation on
the web.

Note that most Field Programmable Gate Arrays (FPGA) are also closely
related to your idea. Take a look at some of the manuals for Xilinx
products (http://www.xilinx.com), for example, and note how look up
tables are used to implement the logic blocks. It is too bad they no
longer make the 6000 series which were the best ones for dynamic
hardware.

Yet another related idea is the BitBLT instruction created by Dan
Ingalls on the old Xerox Alto computer (1973) and which is the heart of
all 2D graphics accelerators. The instruction moved a block of bits
from a source image to a destination image, and its operation was
indicated by 4 bits, one of which was selected to be the new
destination bit depending on the previous value of the destination and
the value of the source bit (exactly your idea). Dan Ingalls showed how
to use this to add together a whole vector of numbers in just a few
instructions with no loop (to implement a cellular automata - the
famous "game of life") in an article in the August 1981 Byte magazine,
so I gave serious consideration to building a machine that had no other
math instructions but this one.

Finally, about if this is the right place to ask about this kind of
thing, looking at the original design for the second generation OC cpu
I would say that it is.

-- Jecel