MATROM.THD  --- Copyright 1988 by Phil Wheeler

An original compilation of Compuserve Model 100 Forum messages for use by
Forum members only.

The Model 100 system bus gives direct access to the innards of the machine,
and has been used to expand it considerably (memory expansions, fast disk
drives, etc.).  This interesting group of messages deals with attatching a
math coprocessor to the system bus.

Message range: 170687 to 171355
Dates: 6/26/88 to 7/8/88


Sb: #Math Rom
Fm: Frank Hausman 71251,2002
To: Stan Wong 70346,1267

I've an Intel 8231 math processor that I'm going to experiment with hooking up
on the system bus.  It does a heap of trig and floating point, and hell of fast
at that!  The other option is the cheapest, slowest 8087, but that would
require an external power supply! I'll let you know if it works...


Fm: Stan Wong 70346,1267
To: Frank Hausman 71251,2002

Wow!  Keep us posted!  I hope to have a "poor man's" ROM development
environment set up on my IBM PC "real soon now."


Fm: Frank Hausman 71251,2002
To: Stan Wong 70346,1267

Well, I've got the circuit designed (not built) and some preliminary code
written. What are you guys most interested in - single precision floating point
(six significant figures), or double precision integer math, or both?  Keep in
mind that memory is a constant restriction. One of the problems so far is that
the chip draws 50-95 milliamperes at 5 volts and (!) 50-95 mA at 12 volts.  And
the 8231/9511 costs about $100.  I haven't completed researching other math
chips, but the 8087 is out since it costs more and seems to depend very heavily
on an 8088 or 8086 being present (and draws 500 mA).
    Two important uses come to mind:  spreadsheets and laboratory data
analysis.  A Forth written with the 8231/9511 in mind would really scream!


Fm: Stan Wong 70346,1267
To: Frank Hausman 71251,2002

Looks like you are really clipping along on this project.  Why don't you drop
Scott T. Schad a line.  I think his interest in your project would be much
greater than mine.

For a math rom, I would think that double precision, and maybe even extended
precision would be a must.  Unless we are using the same terminology different,
the precision should be equal to that of what is commonly found on 16-bit
machines like the IBM PC.  I don't think that people will want to compromise.
They probably would go back to their IBM PC to do such calculations.


Fm: Frank Hausman 71251,2002
To: Stan Wong 70346,1267

The 8231/9511 internally processes 32 bit floating point and integer numbers.
32 bit floating point representations allow about 6 decimal digits of
precision. This is similar to the M100 BASIC's DEFSNG data type.  1.23456E+28
is an example of this data type.
    Six digits may not be enough precision for some folks.  The additional
precision could be obtained in software, at a cost of some calculation speed.
    Another approach is this:  you write all your routines, but leave the
integer multiplies and divides to the 8231.