[TUHS] 68k prototypes & microcode

[John Gilmore]

Clem Cole <clemc at ccc.com> wrote:
> > MC 68K was created in 1980 or thereabouts.

Wikimedia Commons has a pic of a 1979 XC68000L:

  https://commons.wikimedia.org/wiki/File:XC68000.agr.jpg
  https://en.wikipedia.org/wiki/File:XC68000.agr.jpg

After a USENET posting pointed me at them, I browsed the Sunnyvale
Patent Library to bring home the patents for the Motorola 68000.  They
include a full listing of the entire microcode!  I ended up copying it,
taping the sheets together to reconstitute Nick Tredennick's
large-format "hardware flowcharts", and hanging them in the hallway near
my office at Sun.  Fascinating!

I never saw X68000 parts; Sun started in 1981, so Moto had production
parts by then.  But Sun did get early prototypes of the 68010, which we
were very happy for, since we and our customers were running a swapping
Unisoft UNIX because the 68000 couldn't do paging and thus couldn't run
the BSD UNIX that we were porting from the Vax.  Later, I was part of
the Sun bringup team using the XC68020.  We built a big spider-like
daughterboard adapter that would let it be plugged into a 64-pin 68010
socket, so we could debug the 68020 in a Sun-2 CPU board while building
32-bit-wide boards for the Sun-3 bringup.  We had it successfully
running UNIX within a day of receiving it!  (We later heard that our
Moto rep was intending to give that precious early part to another
customer, but decided during their meeting with us to give it to us,
because we were so ready to get it running.)

When the 68000 was announced, it was obviously head-and-shoulders better
than the other clunky 8-bit and 16-bit systems, with a clean 32-bit
architecture and a large address space.  It seems like the designers of
the other chips (e.g. the 8088) had never actually worked with real
computers (mainframes and minicomputers) and kept not-learning from
computing history.

Some of my early experience was in APL implementation on the IBM 360
series.  I knew the 68000 would be a great APL host, since its
autoincrement addressing was perfect for implementing vector operations.
In the process of designing an APL for it (which was never built), I
wrote up a series of short suggestions to Motorola on how to improve the
design.  This was published in Computer Architecture News.  For the
68010 they actually did one of the ideas, the "loop mode" that would
detect a 1-instruction backward decrement-and-branch loop, and stop
continually re-fetching the two instructions.  This made
memory-to-memory or register-vs-memory instruction loops run at almost
the speed of memory, which was a big improvement for bcopy, bzero,
add-up-a-vector-of-integers, etc.

I'll append a USENET posting about the 68000 patents, followed by my
addendum after visiting the Patent office.

	John
	
>From decwrl!decvax!harpo!npoiv!npois!houxm!houxa!houxk!tdl (T.LOVETT) Tue Mar 15 16:55:28 1983
Subject: 68000: 16 bits. With references
Newsgroups: net.micro.68k

With due respect to Henry Spencer I feel that I must correct
some of his statements regarding the 68000. He is correct in
saying that the 68000 is basically 16 bits wide; however,
his explanation of the segmented bus is incorrect. 

The datapath of the 68000 is divided into three pieces, each of
which has two busses, address and data, running through it. Six
busses total. There are muxes which can be switched so that all
address busses are connected and all data busses are connected.
The three sections of the datapath are the data section
(includes low 16 bits of all data registers and ALU), the
"low" section (contains the low 16 bits of address registers and
the low half of the Address Adder(AAU)), and the "high" section
(contains high 16 bits of all address and data registers and
the upper half of the AAU).

Theoretically they could do 6 16 bit transfers simultaneously,
but in looking through the microcode I don't remember seeing more
than three transfers at a time. The "low" and "high" sections can
be cascaded to provide a 32 bit arithmetic unit for address
calculations. 32 bit data calculations must be done in two passes through
the ALU. 

For the masochists out there, you can learn more than you ever wanted
to know about the 68000 by reading Motorola's patents on it. They are
available for some nominal fee (~ one dollar) from the Office
of Patents and Trademarks in Arlington. The relevant patents are:

1 - #4,307,445 "Microprogrammed Control Apparatus Having a Two
	Level Control Store for Data Processor", Tredennick, et al.

	First design of 68000 which was scrapped?

2 - #4,296,469 "Execution Unit for Data Processor using Segmented
	Bus structure", Gunter, et al.

	All about the 16 bit data path

3 - #4,312,034 "ALU and Condition Code Control Unit for Data Processor",
	Gunter, et al.

	Boring.

4 - #4,325,121 "Two-Level Control Store for Microprogrammed Data Processor",
	Gunter et al.

	Bonanza! Full of block diagrams and everything you ever wanted
	to know. Includes complete listing of microcode with
	Tredennick's "hardware flowcharts".

Hope this clears things up.

Tom Lovett BTL Holmdel  harpo!houxk!tdl  201-949-0056


My [gnu] notes on additional 68000 patents:

Pat #		Appl #	Filed date	Issued date	Inventors

4,338,661	041,201	May 21, 1979	Jul 6, 1982
		Tredennick & Gunter
	Conditional Branch Unit for Microprogrammed Data Processor

4,342,078	041,202	May 21, 1979	Jul 27, 1982
		Tredennick & Gunter
	Instruction Register Sequence Decoder for Microprogrammed
	Data Processor and Method

4,312,034	041,203	May 21, 1979	Jan 19, 1982
		Gunter, Hobbs, Spak, Tredennick
	ALU and Condition Code Control Unit for Data Processor

4,325,121	041,135	May 21, 1979	Apr 13, 1982
		Gunter, Tredennick
	Two-Level Control Store for Microprogrammed Data Processor
		Bonanza! Full of block diagrams and everything you ever wanted
		to know. Includes complete listing of microcode with
		Tredennick's "hardware flowcharts".

4,296,469	961,798	Nov 17, 1978	Oct 20, 1981
		Gunter, Tredennick, McAlister
	Execution Unit for Data Processor Using Segmented Bus Structure
		All about the 16 bit data path

4,348,722	136,845	Apr 3, 1980	Sep 7, 1982
		Gunter, Crudele, Zolnowsky, Mothersole
	Bus Error Recognition for Microprogrammed Data Processor

4,349,873	136,593	Apr 2, 1980	Sep 14, 1982
		Gunter, Zolnowsky, Crudele
	Microprocessor Interrupt Processing

4,524,415	447,721	Dec 7, 1982	Jun 18, 1985
		Mills, Moyer, MacGregor, Zolnowsky
	Virtual Machine Data Processor
		68010 changes to 68000

4,348,741	169,558	Jul 17, 1980	Sep 7, 1982
		McAlister, Gunter, Spak, Schriber
	Priority Encoder
		Used to decode the bit masks for MOVEM.

XXXXXXXXX	446,801	Dec 7, 1982
		Crudele, Zolnowsky, Moyer, MacGregor
	Virtual Memory Data Processor

XXXXXXXXX	447,600	Dec 7, 1982
		MacGregor, Moyer, Mills Jr, Zolnowsky
	Data Processor Version Validation
		About how bus errors store a CPU mask version # to
		prevent their being restarted on a different CPU mask
		in a multiprocessor system

XXXXXXXXX	961,796	Nov 17, 1978
		Tredennick et al
	Microprogrammed Control Apparatus for Data Processor
		(continued into 4,325,121, probably never issued)

XXXXXXXXX	961,797	Nov 17, 1978
		McAlister et al
	Multi-port RAM Structure for Data Processor Registers

4,307,445	961,796	Nov 17, 1978
		Tredennick, et al
	Microprogrammed Control Apparatus Having a Two Level
	Control Store for Data Processor
		First design of 68000 which was scrapped?