MemoPET

Introduction

MemoPET is a co-processor board designed for use in MTXPlus+; a project developed by Martin Allcorn and myself to build a "super Memotech MTX". The project is still in the development stage but three working prototypes have been built and running.

 The system architecture uses a multi-slot backplane with separate cards used for the CPU, video display and I/O functions. The core processor is a Z80 but the system supports co-processors installed on their own boards. Martin has already designed a 6502 co-processor board that can be used to emulate a BBC Model B micro.

MemoPET is another 6502 co-processor board, designed to emulate a Commodore PET. Martin describes development of MemoPET below :

Introducing MemoPET

A 6502 based coprocessor designed for the MTXPlus+. (Other than ensuring the MTX PSU has enough power the basic design should also work with the MTX500/512)

Whilst the software is very much work in progress the prototype board has achieved most of the design goals.

The original 6502 co-processor used a minimal design and relies on the MTXPlus+ for all of the input/output, the 6502 can only “see” memory. The new board aims to supply the 6502 with a limited set of peripherals.

The two biggest I/O bottlenecks are the display and disc storage, so those I wanted to put on the board. Keyboard and sound will be passed back to the MTXPlus to deal with. Originally, I wanted to try a large memory system using the 65816 rather than the 6502, however the extra design considerations for accessing the extended memory would have required a completely new OS.

The next option was to go do small memory system like the Acorn Atom or Commodore VIC-20 using the 65816 in 6502 emulation mode. With a patched version of the original OS to allow access to the extra memory banks. However, the video hardware on both systems would have been a struggle to fit into the CPLDs I have available.

The character based display of the older Commodore PET range in not unlike the MTX 80 column display, but without the colour attribute. The VHDL code for that fits the CPLD with room to spare.

Fitting all of the PET components required for the IEEE expansion would have required more space than the Eurocard format allows. However, using the serial IEC port from the Vic and C64 would fit into the space, at the expense of complicating the software.

So, the original design had a 65816 CPU, 512k RAM 512k ROM, along with the external hardware to latch the top 8 bits of the address bus. The video display needed 2k of dual port RAM and an 84 pin CPLD to generate the VGA timing for the 80 column display. The display also needs a ROM with the character definitions . The IEC serial port needs a 6522 VIA.

Since I wanted to try and emulate the 40 column pet display as well as the larger one, the shift register, colour register and colour multiplexer are all external components. This leaves the maximum available space in the CPLD for logic.

VGA at 640x480 specifies a 25.175MHz clock so the board needs an oscillator, most monitors will lock on to VGA generated from a 25MHz clock. Since that’s not available from the MTXPlus+ either, I fitted the correct speed part. That needs dividing by 2 to generate the clock for the 40 column display, halving that again would produce a clock just under 6.3mhz that I was going to try and run the CPU from.

Communication with the MTXPlus+ required some way to transmit between the 2 CPU’s Rather than use another 48 pin (and expensive) dual port ram I decided to use two pairs of 4 bit registers to allow for 4 byte data transfers in each direction. Mapping that into the Z80’s I/O map needed a small CPLD even if there’s available space, and at the time I wasn’t sure there would be, there aren’t enough pins on the big CPLD.

It took several attempt, but eventually I had a design that would fit the 19 chips and other parts into the available space.

The Fist 64k of the memory map was designed to be as close to a real pet as possible.

Full 1MB Memory Ma
FFFFF F0000	Data RAM
E0000	Data RAM
D0000	Data RAM
C0000	Data RAM
B0000	Data RAM
A0000	Data RAM
90000	Data RAM
88000 80000	Data RAM Low Memory RAM
70000	Data ROM
60000	Data ROM
50000	Data ROM
40000	Data ROM
30000	Data ROM
20000	Data ROM
10000	Data ROM
00000	Low Memory Block

The extended memory map was kept simple, and split evenly between ROM and RAM

Initial tests showed issues either from the wiring of memory, or the memory paging arrangement. In order to sort determine where the issues were, the board was re-wired to take the current production version of the 6502, which is the W65C02S.

Dispensing with the 65816 allowed for the removal of the 2 devices involved in separating the data bus and upper address bus. It also allowed for the fitting of a physically smaller 32k ram. A smaller 28 pin eeprom could have been used but they’re usually slower than the 32 pin flash devices. The prototype still has a 32pin device, but only portions of the first 32k are used.

That leaves the board with a just the low part of the memory map, very much like the original pet. The major differences are in the I/O area.

Re-wiring for the 6502 helped identify an issue with the address bus preventing the system working. At which point I had the basic components CPU, ROM & RAM connected and working.

Adapting the VHDL display code from the CFX-3 to produce a mono display with 40 and 80 column options is more or less complete, allowing for further testing of the rest of the board. The extra logic to allow for the separate display opting turned out to only need a couple of extra macro-cells, however the number of pins in use means incorporating the external video components isn’t an option.

The 6502 chip like the Z80 has 2 interrupt pins IRQ that can be turned off in software, and NMI that cannot. The PET Kernel uses the NMI to trigger the break process and return to the system prompt. In order for the Z80 side, which will provide the keyboard support to be able to support this, a extra pin from the second logic device has been programmed to issue a short pulse when a designated I/O port is accessed.

The other interrupt supports both the 50/60hz system counter and the 6522 VIA, so the interrupt out from the VIA has to be connected to the CPLD where it can be mixed with vsync and passed to the CPU.

2 way communication with the Z80 CPU to provide keyboard input has been incorporated into a modified pet “Editor” rom. The Z80 side is currently a short basic program that scans the keyboard on request. That’s enough to type in and run short programs in Commodore Basic 4 on the MemoPET board.

The board itself beats expectations for the speed at which it will run at. It has shown no signs of instability when running off the 12.6mhz clock intended for the 40 column VGA display instead of the 6.3 clock I had intended to use. Part of this may be the decision to use the 6502 in place of the 65816 so for now there are no plans to revert back to the bigger chip, the Memopet will retain the older CPU and smaller memory map.

On thing that hasn’t be successful so far is the IEC interface, the timing for the serial interface is entirely done in software and to date I’ve not managed to get the board to connect to any IEC peripherals.

To try and get to the bottom of the, I’ve built an additional clock board that plus into the sockets freed up by the 65816 to 6502 re-wire. That clock runs at the same frequency as the clock in the Vic20 so while further code testing takes place the IEC code from the Vic can be used without needing to “tweak” the timing delays.

Merging the IEC code into the Pet ROMs still needs to be done and that is the next task. That needs more investigations of exactly how the Pet handles the IEEE interface so could take some time.  

 

 

An early version of the schematic that Martin produced before starting to wire the prototype - as usual, Martin used Excel to do the layout, the components are viewed from the solder side. Click on the image to see a PDF with more detail
First build of the prototype board, with the sockets placed, ready for wiring. It may be of interest to note Martin's preference for single sided boards. He feels that they are easier to solder, though less heat tolerant than double sided proto boards.
Solder side of the board at the same stage
Now for the hard work - wiring the board ! Power, Ground and Clock (white wires) signals have been connected. This is actually a disproportionate amount of work than the number of wires would suggest. Power and Ground require multiple terminations, whereas many of the signal wires are just point-to-point connections.
Now with most of the video signal wiring done . . . The CPLD will be responsible for generating a 10 or 11 bit video address to the VRAM. (blue wires) The VRAM data output (green) along with 3 bit row counter from the CPLD (white) and one bit from the VIA produces a 12 bit address to the character rom. The output from the ROM (yellow) goes to the shift register next door. The colour (RGB coloured) data comes from an 8 bit latch with 2 bits unused.
The foreground colour data is inverted before foreground and background data was to the multiplexer driven by the shift register output. Inverting the foreground colour from the latch will ensure a default output of white on black. In this slide, the majority of the wiring is done, just the low half of the address bus and the data bus need to be completed, but as you can see, the wiring around the CPLD is pretty conjested.
Phew ! - The MemoPET board is now fully wired. Martin has made a couple of changes to the design along the way . . . . The CPLD now has all 3 clocks fed to it. The shift register isn't connected directly to the VGA clock any more, that's connected to a spare CPLD pin.
That means that, theoretically, the CPLD can produce a 40 column display using the intermediate half VGA clock, as well as 80 columns from the standard clock.  The 1/4 speed clock drives the CPU and VIA at 6.3mhs so the CPLD needs that for the read/write logic. Reflecting these changes, the schematic has been updated
With the wiring completed, here you can see the first video output. If you open up the larger image file, you can see that there are characters being produced from the Vic-20 character set. Obviously, there are issues such as the character colour (which should be white, not violet) and the matrix size, but this is a great start.
After a little "tweaking" of the software, a more recognisable PET like display. The display is inverted, but that will be resolved by selecting the other output from the shift register. The display is currently missing a small portion of the left hand character, but that will be resolved by adjusting the video timing slightly.
Video inversion fixed and screen alignment almost resolved. In the next two images, you can see that the screen is correctly displaying 40 characters.
But if you look very closely, you may be able to see that the first pixel of each character is a little wider than it should be, while the last pixel is a little narrower. This will be fixed by adding an R-C delay, rather than messing with the CPLD code.
The assembled board, plugged into the MTXPlus+ backplane. The two unpopulated sockets are no longer needed and the ZIF socket is just installed for ease of testing ROM updates. Now that the display system is pretty much working, the next, and bigger, challenge, is to get the disk I/O interface working.
Updated schematic - open the PDF to for some brief details of how the board PIA and VIA work in comparison to a PET or VIC-20.
A little taster . . . The 20k, 40 column PET ROM image into loaded into the flash in the MemoPET.   Obviously, there's no support for anything other than the screen memory, but it's a promising start!
This image shows the PET RAM being written to from the MTXPlus+. A BASIC program on the MTXPlus+ side writes data to a 4 digit counter and triggers an NMI interrupt on the 6502 side. The MemoPET output shows the values of the four digit counter on the second line and the value of the interrupt counter on the fourth line.
Now looking remarkably "PET like" ! Using a modified Editor ROM, created with some help from Steve Gray's PET Editor ROM pages, and adding elements from the PET BASIC 4 and Kernel ROMs things are starting to come together nicely. At this stage, there were some issues with key-mapping and keyboard de-bounce, but things are obviously progressing . . . .
Now demonstrating a running Commodore PET BASIC program. The program was printing numbers that were counting down under IRQ control, derived from the vertical blank interrupt. The program has been stopped in a similar way that a "real" PET BASIC program would be stopped with the Run/Stop Key. Still to be completed are the I/O routines that can drive simple sound output and an emulated disk drive (IEEE-488/IEC) interface like the PET, or more accurately, the VIC-20.