MTX Co-Processor

Courtesy of Martin Allcorn

Martin has developed a co-processor interface for the MTX, inspired by the "Tube" interface of the BBC Micro, he's calling it MTXPipe^TM.

The tube was a fast bus interface that provided connectivity to additional co-processors, including a  faster 6502, a Z80 or an ARM processor, the BBC Micro continued to handle I/O processing whilst the additional co-processor could execute the Language Application.

Acorn co-processors used a custom ULA in the co-processor expansion "wedge" that handled the inter-processor communication, the processors were not synchronised, the ULA included buffers used to store data until the other processor was ready to read it.

Design & Build
Martin's mock up of his initial co-processor interface design. The 74HCT245 transceivers make up the status port for each CPU to read the FIFO status, 245s were chosen for their tri-state outputs. The 4, 16 pin chips are 4 bit wide, 16 level FIFO memories, they're paired up to make the 8 bit wide send/receive pipes between the CPU's. The 74HCT74 dual flip flop is an interrupt controller, one CPU operates the set function, the other drives the reset.  The 24 pin GAL is on the MTX side, the 20 pin on the co-pro side, they do all the timing signals for the FIFOs. The MTX side does a full I/O decode which is why it needs the extra pins. The Co-pro side will need logic on the CPU board to make the signals more Z80 like, so the interface only needs to do a partial decode.
The proposed design MTXpipe uses 4 addresses/ports to :     Read/write the FIFOs      Read the status register     Set the interrupt flip-flop     Reset the send FIFO      Reset the write FIFO     Clear the interrupt flip-flop       Reset both FIFOs        Using reads or writes to perform the reset/interrupt actions instead of using bits in a register saves board space.   Both sides would get the same 4 port access, at this point, the intention was that the MTX side would be passive, and responds to requests from the Co-pro, which is effectively in charge. The plan was that the software should put the relevant transaction data in the pipe then set the interrupt. 16 bytes is ample for screen/keyboard processing. Genpat is the longest VDU sequence at 12 bytes.
Martin then developed a simplified design using a 1 kB Dual Port RAM, the interrupt flip flops and data registers are no longer required and the paging logic is much simpler, being memory mapped instead of I/O mapped. Each “side” gets its own read/write/chip select pins and a BUSY signal is wired to the CPU’s WAIT pin to stop simultaneous access to the same location.  One location on each side sends an interrupt to the other side, reading the location from the other side resets the interrupt.
The initial test passes with flying colours, the ROM pages in and out as required. The next test will be to plug in the 1kB dual port RAM and write some code to test it on the Z80 side.
The MTXPipe board connected to a very basic 6502 Co-processor board, temporarily using ZIF sockets for ease of replacement during testing. The co-processor board has 32kB RAM, 32kB of ROM, it is clocked of the MTX 4MHz clock signal but will ultimately have its own, faster, CPU clock.
The simple test ROM on the 6502 side writes a test pattern into the shared RAM, and the Z80 side is able to read it back.
The system booting from a simple ROM on the Co-processor board, as the screen shows, there is a small issue with the displayed text - an extra character in the "Ready" text, possibly related to a timing issue that required further investigation.
Some progress, text can be typed on the MTX keyboard and passed over to the 6502 processor, although it is being misinterpreted on the 6502 side - even though it is displayed correctly. The initial goal is to get BBC BASIC to run on the system, so far, Operating System calls OSWRCH, OSRDCH and minimal OSWORD and OSBYTE are supported.
Design development . . . . The Co-processor now has it's own clock, the 12MHz oscillator is directly driving the CPU without the need for any supporting components. The additional resistor adjacent to the connector is required to prevent the CPU from affecting the interrupt line during a WAit-for-Interrupt (WAI) instruction which was added in the WDC W65C02 processor.
The system now reporting 46K of RAM and running BBC BASIC 3. PCW benchmark 8 (which performed maths operations) speeds :         49.9 seconds for the BBC Model B (2 MHz 6502)         44.9 seconds for the MTX 512 (4MHz Z80)         8.2 seconds for the Co-processor (12 MHz 6502)
Now with limited support for VDU codes that are compatible with the MTX’s display system, so VDU 26 is “home” and not VDU 30 which the BBC uses. Likewise COLOUR in basic will do weird things, but VDU 4,x or VDU 6,x will set the colours.
To compare the speed of computers for their reviews in the 80s, Personal Computer World used a series of eight BASIC benchmark programs. Results for the Co-pro are shown in the table, they have a mean of 1.87s, compared to the BBC Micros 14.3s and the MTX500 of 19.3s	BM1  0.09s BM2  0.38s BM3  1.06s  BM4  1.11s BM5  1.17s  BM6  1.79s  BM7  2.77s  BM8  6.61s	(Click to see benchmark details)
The co-processor board with a form factor suitable for connection to the MTXPlus+ system bus. The Z80 is clocked on the MTXPlus+ side at 10MHz and the 6502 is clocked at 16MHz by the on board oscillator. The large (48-pin) DIP chip is the dual-port RAM, rather than the PLCC used in the MTX format board. Footprint is not an issue here and the DIP was easier to wire than a PLCC socket.
Wiring side of the MTXPlus+ board, the wiring for the dual port RAM clearly shows the segregation between the MTX side and the 6502 side. The only signal to cross domains is RESET, the 6502 system is totally self contained.
The MTXPlus+ co-pro running BBC BASIC and loading a program from the MTXPlus+ Compact Flash drive
December 2015 I received a surprise early Christmas present from Martin ! Wow ! - One fully built 6502 Co-Processor board for MTXPlus+
Annotated to show the major components on the board. The largest chip is the 1k x 8, dual port Static RAM, an IDT 7130
A higher res. photo of the wire side of "my" new co-processor board. As usual, I am really impressed by the quality of Martin's construction - my boards never look that neat !

MTX Version Schematic
Schematic of the current version of the MTX Co-processor board. Although the schematic shows a 128kb EEROM and a 128 kb RAM on the 6502 side, other devices are supported, including flash memory and other sizes of RAM.

The Techie Bit

With the design pretty much complete, it's time to say a few more words about how it works. Martin has written a few words of explanation in his post on Memorum.

Currently, there are two different versions of the co-pro board, one for the original MTX and one for MTXPlus+. The MTX version is built in 2 parts, an Interface board and the processor board.

It's done that way for 2 reasons:

1. It leaves more room on the processor board for components - this was not an issue with the test board, but it could be for other processors like the 68008 or 9900 that Lez generously donated. These are physical larger and in the 9900's case, have a 16 bit data bus.

2. Cost. The 1K dual port ram used for communications actually cost [Martin] more than the 6502 did

Interface Board

The slim interface board consists of an 8k ROM, the dual port RAM and a 15ns GAL.

The GAL pages the ROM into slot 7 and is used to save space, timing on the MTX side is fairly relaxed and so I could have used standard logic chips. The GAL also pages the top 256 bytes of the 1k RAM into the top of the ROM bank. [Martin] used the Cypress CY7C-131 on this board, and the IDT7130 on the MTXPlus+ version of the board as they both work the same way.

The top part on the RAM is required, as the top 2 locations of ram also act as interrupt controllers, one for each side. 256 bytes are used as this enables the transfer of a full CPM sector and control information, with the minimal loss of available ROM space.

6502 Processor Board

The 6502 processor board has a W65C02-14, 128k static RAM, 128k flash, a 7ns GAL and an oscillator to allow the 6502 to be clocked independently from the Z80.

For a language, the 6502 is currently using HiBasic 3 from the BBC Micro's external 6502 processor, that's a 16k image, and is designed to allow 2k space for the Tube OS, I've used that space, and allocated another 2k below the language rom, making 20k in total. That leaves room for 44k of system ram. The GAL pages the RAM chip in for the lower 44k of the memory map, and the ROM in for most of the upper 20k. The 256 bytes of the dual port RAM is fitted into a "page" near the top of the 2k tube ROM area. The un-used areas of the Flash and RAM aren't accessible

The 6502 is currently clocked at 14.3Mhz (16MHz on the MTXPlus+ version), which is why the board needs the fast GAL, standard logic would have restricted the CPU to a much slower clock rate. The 6502 is in charge of all communications across the dual port RAM. The replacement for the tube ROM has a VERY restricted subset of the BBC MOS running, but it is enough to print, read the keyboard, and save/load basic to the CFX that's plugged into the external connector on the other end of the MTX..

The Z80 however controls the 6502 interrupts, and is currently generating a 100hz interrupt as used by the BBC. The 6502 to Z80 interrupt isn't connected. The Z80 control software works by constantly reading the shared ram for new instructions from the 6502.

References :

Support Group Application Note AN004, Acorn Computers, 1992

BBC BASIC On-line reference manual at benryves.com

BeebWiki - 8-bit Acorn Computer Wiki on mdfs.net