| Project | Forth SoC written in VHDL |
|---|---|
| Author | Richard James Howe |
| Copyright | 2013-2019 Richard Howe |
| License | MIT/LGPL |
| howe.r.j.89@gmail.com |
This project implements a small stack computer tailored to executing Forth based on the J1 CPU. The processor has been rewritten in VHDL from Verilog, and extended slightly.
The goals of the project are as follows:
- Create a working version of J1 processor (called the H2).
- Make a working toolchain for the processor.
- Create a FORTH for the processor which can take its input either from a UART or a USB keyboard and a VGA adapter.
All three of which have been completed.
The H2 processor, like the J1, is a stack based processor that executes an instruction set especially suited for FORTH.
The current target is the Nexys3 board, with a Xilinx Spartan-6 XC6LX16-CS324 FPGA, new boards will be targeted in the future as this board is reaching it's end of life. The VHDL is written in a generic way, with hardware components being inferred instead of explicitly instantiated, this should make the code fairly portable, although the interfaces to the Nexys3 board components are specific to the peripherals on that board.
A video of the project in action, on the hardware, can be viewed here:
demo.mp4
The SoC can also be simulated with a simulator written in C, as shown below:
The System Architecture is as follows:
The licenses used by the project are mixed and are on a per file basis. For my code I use the MIT license - so feel free to use it as you wish. The other licenses used are the LGPL and the Apache 2.0 license, they are confined to single modules so could be removed if you have some aversion to LGPL code.
The only target board available at the moment is the Nexys3, this should change in the future as the board is currently at it's End Of Life. The next boards I am looking to support are it's successor, the Nexys 4, and the myStorm BlackIce (https://mystorm.uk/). The myStorm board uses a completely open source toolchain for synthesis, place and route and bit file generation.
The build has been tested under Debian Linux, version 8.
You will require:
- GCC, or a suitable C compiler capable of compiling C99
- Make
- Xilinx ISE version 14.7
- GHDL
- GTKWave
- tcl version 8.6
- Digilent Adept2 runtime and Digilent Adept2 utilities available at http://store.digilentinc.com/digilent-adept-2-download-only/
- freeglut (for the GUI simulator only)
- pandoc for building the documentation
- picocom (or an alternative terminal client)
Hardware:
- VGA Monitor, and cable (Optional)
- USB Keyboard (Optional) (plugs into the Nexys3 USB to PS/2 bridge)
- Nexys3 development board (if communication via UART only is desired, the VGA Monitor and USB and Keyboard are not needed).
- USB Cables!
Xilinx ISE can (or could be) downloaded for free, but requires registration. ISE needs to be on your path:
PATH=$PATH:/opt/Xilinx/14.7/ISE_DS/ISE/bin/lin64;
PATH=$PATH:/opt/Xilinx/14.7/ISE_DS/ISE/lib/lin64;
To make the C based toolchain:
make embed.hex
To make a bit file that can be flashed to the target board:
make simulation synthesis implementation bitfile
To upload the bitfile to the target board:
make upload
To view the wave form generated by "make simulation":
make viewer
The C based CLI simulator can be invoked with:
make run
Which will assemble the H2 Forth source file embed.fth, and run the assembled object file under the H2 simulator with the debugger activated. A graphical simulator can be run with:
make gui-run
Which requires freeglut as well as a C compiler.
The original J1 project is available at:
This project targets the original J1 core and provides a eForth implementation (written using Gforth as for meta-compilation/cross compilation to the J1 core). It also provides a simulator for the system written in C.
The eForth interpreter which the meta-compiler is built on can be found at:
The H2 processor and associated peripherals are now quite stable, however the source is always the definitive guide as to how instructions and peripherals behave, as well as the register map.
There are a few modifications to the J1 CPU which include:
- New instructions
- A CPU hold line which keeps the processor in the same state so long as it is high.
- Interrupt Service Routines have been added.
- Larger (adjustable at time of synthesis) return and data stacks
The H2 CPU behaves very similarly to the J1 CPU, and the J1 PDF can be read in order to better understand this processor. The processor is 16-bit with instructions taking a single clock cycle. Most of the primitive Forth words can also be executed in a single cycle as well, one notable exception is store ("!"), which is split into two instructions.
The CPU has the following state within it:
- A 64 deep return stack (up from 32 in the original J1)
- A 65 deep variable stack (up from 33 in the original J1)
- A program counter
- An interrupt enable and interrupt request bit
- An interrupt address register
- Registers to delay and hold the latest IRQ and hold-line values
Loads and stores into the block RAM that holds the H2 program discard the lowest bit, every other memory operation uses the lower bit (such as jumps and loads and stores to Input/Output peripherals). This is so applications can use the lowest bit for character operations when accessing the program RAM.
The instruction set is decoded in the following manner:
+---------------------------------------------------------------+
| F | E | D | C | B | A | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+---------------------------------------------------------------+
| 1 | LITERAL VALUE |
+---------------------------------------------------------------+
| 0 | 0 | 0 | BRANCH TARGET ADDRESS |
+---------------------------------------------------------------+
| 0 | 0 | 1 | CONDITIONAL BRANCH TARGET ADDRESS |
+---------------------------------------------------------------+
| 0 | 1 | 0 | CALL TARGET ADDRESS |
+---------------------------------------------------------------+
| 0 | 1 | 1 | ALU OPERATION |T2N|T2R|N2A|R2P| RSTACK| DSTACK|
+---------------------------------------------------------------+
| F | E | D | C | B | A | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+---------------------------------------------------------------+
T : Top of data stack
N : Next on data stack
PC : Program Counter
LITERAL VALUES : push a value onto the data stack
CONDITIONAL : BRANCHS pop and test the T
CALLS : PC+1 onto the return stack
T2N : Move T to N
T2R : Move T to top of return stack
N2A : STORE T to memory location addressed by N
R2P : Move top of return stack to PC
RSTACK and DSTACK are signed values (twos compliment) that are
the stack delta (the amount to increment or decrement the stack
by for their respective stacks: return and data)
All ALU operations replace T:
| Value | Operation | Description |
|---|---|---|
| 0 | T | Top of Stack |
| 1 | N | Copy T to N |
| 2 | T + N | Addition |
| 3 | T & N | Bitwise AND |
| 4 | T or N | Bitwise OR |
| 5 | T ^ N | Bitwise XOR |
| 6 | ~T | Bitwise Inversion |
| 7 | T = N | Equality test |
| 8 | N < T | Signed comparison |
| 9 | N >> T | Logical Right Shift |
| 10 | T - 1 | Decrement |
| 11 | R | Top of return stack |
| 12 | [T] | Load from address |
| 13 | N << T | Logical Left Shift |
| 14 | depth | Depth of stack |
| 15 | N u< T | Unsigned comparison |
| 16 | Set CPU State | Enable interrupts |
| 17 | Get CPU State | Are interrupts on? |
| 18 | rdepth | Depth of return stk |
| 19 | 0= | T == 0? |
| 20 | CPU ID | CPU Identifier |
| 21 | LITERAL | Internal Instruction |
Registers marked prefixed with an 'o' are output registers, those with an 'i' prefix are input registers. Registers are divided into an input and output section of registers and the addresses of the input and output registers do not correspond to each other in all cases.
The following peripherals have been implemented in the VHDL SoC to interface with devices on the Nexys3 board:
- VGA output device, text mode only, 80 by 40 characters from
http://www.javiervalcarce.eu/html/vhdl-vga80x40-en.html. This has
been heavily modified from the original, which now implements most of a
VT100 terminal emulator. This has two fonts available to it:
- Terminus/KOI8-R (Default)
- Latin ISO-8859-15 (Secondary Font) from https://git.kernel.org/pub/scm/linux/kernel/git/legion/kbd.git
- [Timer][] in timer.vhd.
- UART (Rx/Tx) in uart.vhd.
- PS/2 Keyboard from https://eewiki.net/pages/viewpage.action?pageId=28279002
- LED next to a bank of switches
- A 7 Segment LED Display driver (a 7 segment display with a decimal point)
The SoC also features a limited set of interrupts that can be enabled or disabled.
The output register map:
| Register | Address | Description |
|---|---|---|
| oUart | 0x4000 | UART register |
| oVT100 | 0x4002 | VT100 Terminal Write |
| oLeds | 0x4004 | LED outputs |
| oTimerCtrl | 0x4006 | Timer control |
| oMemDout | 0x4008 | Memory Data Output |
| oMemControl | 0x400A | Memory Control / Hi Address |
| oMemAddrLow | 0x400C | Memory Lo Address |
| o7SegLED | 0x400E | 4 x LED 7 Segment display |
| oIrcMask | 0x4010 | CPU Interrupt Mask |
| oUartBaudTx | 0x4012 | UART Tx Baud Clock Setting |
| oUartBaudRx | 0x4014 | UART Rx Baud Clock Setting |
The input registers:
| Register | Address | Description |
|---|---|---|
| iUart | 0x4000 | UART register |
| iVT100 | 0x4002 | Terminal status & PS/2 Keyboard |
| iSwitches | 0x4004 | Buttons and switches |
| iTimerDin | 0x4006 | Current Timer Value |
| iMemDin | 0x4008 | Memory Data Input |
The following description of the registers should be read in order and describe how the peripherals work as well.
A UART with a fixed baud rate and format (115200, 8 bits, 1 stop bit) is present on the SoC. The UART has a FIFO of depth 8 on both the RX and TX channels. The control of the UART is split across oUart and iUart.
To write a value to the UART assert TXWE along with putting the data in TXDO. The FIFO state can be analyzed by looking at the iUart register.
To read a value from the UART: iUart can be checked to see if data is present in the FIFO, if it is assert RXRE in the oUart register, on the next clock cycle the data will be present in the iUart register.
The baud rate of the UART can be changed by rebuilding the VHDL project, bit length, parity bits and stop bits can only be changed with modifications to uart.vhd
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| X | X |TXWE| X | X |RXRE| X | X | TXDO |
+-------------------------------------------------------------------------------+
TXWE: UART TX Write Enable
RXRE: UART RX Read Enable
TXDO: UART TX Data Output
The VGA Text device emulates a terminal which the user can talk to by writing to the oVT100 register. It supports a subset of the VT100 terminal functionality. The interface behaves much like writing to a UART with the same busy and control signals. The input is taken from a PS/2 keyboard available on the board, this behaves like the RX mechanism of the UART.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| X | X |TXWE| X | X |RXRE| X | X | TXDO |
+-------------------------------------------------------------------------------+
TXWE: VT100 TX Write Enable
RXRE: UART RX Read Enable
TXDO: UART TX Data Output
On the Nexys3 board there is a bank of LEDs that are situated next to the switches, these LEDs can be turned on (1) or off (0) by writing to LEDO. Each LED here corresponds to the switch it is next to.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| X | X | X | X | X | X | X | X | LEDO |
+-------------------------------------------------------------------------------+
LEDO: LED Output
The timer is controllable by the oTimerCtrl register, it is a 13-bit timer running at 100MHz, it can optionally generate interrupts and the current timers internal count can be read back in with the iTimerDin register.
The timer counts once the TE bit is asserted, once the timer reaches TCMP value it wraps around and can optionally generate an interrupt by asserting INTE. This also toggles the Q and NQ lines that come out of the timer and are routed to pins on the board (see the constraints file top.ucf for the pins).
The timer can be reset by writing to RST.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| TE | RST|INTE| TCMP |
+-------------------------------------------------------------------------------+
TE: Timer Enable
RST: Timer Reset
INTE: Interrupt Enable
TCMP: Timer Compare Value
The H2 core has a mechanism for interrupts, interrupts have to be enabled or disabled with an instruction. Each interrupt can be masked off with a bit in IMSK to enable that specific interrupt. A '1' in a bit of IMSK enables that specific interrupt, which will be delivered to the CPU if interrupts are enabled within it.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| X | X | X | X | X | X | X | X | IMSK |
+-------------------------------------------------------------------------------+
IMSK: Interrupt Mask
This register is used to set the baud and sample clock frequency for transmission only.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| BTXC |
+-------------------------------------------------------------------------------+
BTXC: Baud Clock Settings
This register is used to set the baud and sample clock frequency for reception only.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| BRXC |
+-------------------------------------------------------------------------------+
BRXC: Baud Clock Settings
Data to be output to selected address when write enable (WE) issued in oMemControl.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| Data Ouput |
+-------------------------------------------------------------------------------+
This register contains the control registers for the onboard memory on the Nexys3 board. The board contains three memory devices, two non-volatile memory devices and a volatile RAM based device. The two devices accessible by a simple SRAM interface (one volatile M45W8MW16, one non-volatile - a NP8P128A13T1760E) are both accessible, the third is an SPI based memory device, NP5Q128A13ESFC0E) and is currently not accessible.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| OE | WE | RST|WAIT| RCS| FCS| Address Hi |
+-------------------------------------------------------------------------------+
OE: Output Enable - enable reading from current address into iMemDin
WE: Write Enable - enable writing oMemDout into ram at current address
RST: Reset the Flash memory controller
RCS: RAM Chip Select, Enable Volatile Memory
FCS: Flash Chip Select, Enable Non-Volatile Memory
Address Hi: High Bits of RAM address
OE and WE are mutually exclusive, if both are set then there is no effect.
The memory controller is in active development, and the interface to it might change.
This is the lower address bits of the RAM.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| Address Lo |
+-------------------------------------------------------------------------------+
On the Nexys3 board there is a bank of 7 segment displays, with a decimal point (8-segment really), which can be used for numeric output. The LED segments cannot be directly addressed. Instead the value stored in L8SD is mapped to a hexadecimal display value (or a BCD value, but this requires regeneration of the SoC and modification of a generic in the VHDL).
The value '0' corresponds to a zero displayed on the LED segment, '15' to an 'F', etcetera.
There are 4 displays in a row.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| L7SD0 | L7SD1 | L7SD2 | L7SD3 |
+-------------------------------------------------------------------------------+
L7SD0: LED 7 Segment Display (leftmost display)
L7SD1: LED 7 Segment Display
L7SD2: LED 7 Segment Display
L7SD3: LED 7 Segment Display (right most display)
The iUart register works in conjunction with the oUart register. The status of the FIFO that buffers both transmission and reception of bytes is available in the iUart register, as well as any received bytes.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| X | X | X |TFFL|TFEM| X |RFFL|RFEM| RXDI |
+-------------------------------------------------------------------------------+
TFFL: UART TX FIFO Full
TFEM: UART TX FIFO Empty
RFFL: UART RX FIFO Full
RFEM: UART RX FIFO Empty
RXDI: UART RX Data Input
The iVT100 register works in conjunction with the oVT100 register. The status of the FIFO that buffers both transmission and reception of bytes is available in the iVT100 register, as well as any received bytes. It works the same as the iUart/oUart registers.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| X | X | X |TFFL|TFEM| X |RFFL|RFEM| 0 | ACHR |
+-------------------------------------------------------------------------------+
TFFL: VGA VT100 TX FIFO Full
TFEM: VGA VT100 TX FIFO Empty
RFFL: PS2 VT100 RX FIFO Full
RFEM: PS2 VT100 RX FIFO Empty
ACHR: New character available on PS2 Keyboard
This register contains the current value of the timers counter.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| X | X | X | TCNT |
+-------------------------------------------------------------------------------+
TCNT: Timer Counter Value
iSwitches contains input lines from multiple sources. The buttons (BUP, BDWN, BLFT, BRGH, and BCNT) correspond to a D-Pad on the Nexys3 board. The switches (TSWI) are the ones mentioned in oLeds, each have an LED next to them.
The switches and the buttons are already debounced in hardware so they do not have to be further processed once read in from these registers.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| X | X | X | BUP|BDWN|BLFT|BRGH|BCNT| TSWI |
+-------------------------------------------------------------------------------+
BUP: Button Up
BDWN: Button Down
BLFT: Button Left
BRGH: Button Right
BCNT: Button Center
TSWI: Two Position Switches
Memory input, either from the SRAM or Flash, indexed by oMemControl and oMemAddrLow. When reading from flash this might actually be status information or information from the query table.
+-------------------------------------------------------------------------------+
| 15 | 14 | 13 | 12 | 11 | 10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
+-------------------------------------------------------------------------------+
| Data Input |
+-------------------------------------------------------------------------------+
The following interrupt service routines are defined:
| Name | Number | Description |
|---|---|---|
| isrNone | 0 | Not used |
| isrRxFifoNotEmpty | 1 | UART RX FIFO Is Not Empty |
| isrRxFifoFull | 2 | UART RX FIFI Is Full |
| isrTxFifoNotEmpty | 3 | UART TX FIFO Is Not Empty |
| isrTxFifoFull | 4 | UART TX FIFO Is Full |
| isrKbdNew | 5 | New PS/2 Keyboard Character |
| isrTimer | 6 | Timer Counter |
| isrDPadButton | 7 | Any D-Pad Button Change State |
When an interrupt occurs, and interrupts are enabled within the processor, then a call to the location in memory is performed - the location is the same as the ISR number. An ISR with a number of '4' will perform a call (not a jump) to the location '4' within memory, for example.
Interrupts have a latency of at least 4-5 cycles before they are acted on, there is a two to three cycle delay in the interrupt request handler, then the call to the ISR location in memory has to be done, then the call to the word that implements the ISR itself.
If two interrupts occur at the same time they are processed from the lowest interrupt number to the highest.
Interrupts are lost when an interrupt with the same number occurs that has not been processed.
The Disassembler and C based simulator for the H2 is in a single program (see h2.c). This simulator complements the VHDL test bench tb.vhd and is not a replacement for it. The meta-compiler runs on top of an eForth interpreter and it contained within the files embed.c and embed.blk. The meta-compiler (Forth parlance for a cross-compiler) is a Forth program which is used to create the eForth image that runs on the target.
The toolchain is currently in flux, going forward there is liable to more integration between h2.c and embed.c, along with changing the Embed Virtual Machine into one that more closely resembles the H2 CPU with the long term goal of creating a self hosting system.
To build both, a C compiler is needed, the build target "h2" will build the executable, h2, and "embed" will build the meta-compiler:
make h2 embed
And it can be run on the source file embed.fth with the make target:
make run
The make file is not needed:
Linux:
cc -std=c99 h2.c -o h2 # To build the h2 executable
cc -std=c99 embed.c -o embed # To build the embed VM executable
./embed embed.blk embed.hex embed.fth # Create the target eForth image
./h2 -h # For a list of options
./h2 -r embed.hex # Run the assembled file
Windows:
gcc -std=c99 h2.c -o h2.exe # Builds the h2.exe executable
gcc -std=c99 embed.c -o embed.exe # Builds the embed.exe executable
embed.exe embed.blk embed.hex embed.fth # Create the target eForth iamge
h2.exe -h # For a list of options
h2.exe -r embed.hex # Run the assembled file
A list of command line options available:
- stop processing options, following arguments are files
-h print a help message and exit
-v increase logging level
-d disassemble input files (default)
-D full disassembly of input files
-T Enter debug mode when running simulation
-r run hex file
-L # load symbol file
-s # number of steps to run simulation (0 = forever)
-n # specify NVRAM block file (default is nvram.blk)
file* file to process
This program is released under the MIT license, feel free to use it and modify it as you please. With minimal modification it should be able to assemble programs for the original J1 core.
The meta-compiler runs on top of the embed virtual machine, it is a 16-bit virtual machine that originally descended from the H2 CPU. The project includes a meta-compilation scheme that allows an eForth image to generate a new eForth image with modifications. That system has been adapted for use with the H2, which replaced the cross compiler written in C, which allowed the first image for the H2 to be created.
The meta-compiler is an ordinary Forth program, it is contained within embed.fth. The meta-compiler Forth program is then used to build up an eForth image capable of running on the H2 target.
For more information about meta-compilation in Forth, see:
- https://github.com/howerj/embed
- https://github.com/howerj/embed/blob/master/embed.fth
- http://www.ultratechnology.com/meta.html
The disassembler takes a text file containing the assembled program, which consists of 16-bit hexadecimal numbers. It then attempts to disassemble the instructions. It can also be fed a symbols file which can be generated by the assembler and attempt to find the locations jumps and calls point to.
The disassembler is used by a tcl script called by GTKwave, it turns the instruction trace of the H2 from a series of numbers into the instructions and branch destinations that they represent. This makes debugging the VHDL much easier.
The purple trace shows the disassembled instructions.
The simulator in C implements the H2 core and most of the SoC. The IO for the simulator is not cycle accurate, but can be used for running and debugging programs with results that are very similar to how the hardware behaves. This is much faster than rebuilding the bit file used to flash the FPGA.
The simulator also includes a debugger, which is designed to be similar to the DEBUG.COM program available in DOS. The debugger can be used to disassemble sections of memory, inspect the status of the peripherals and dump sections of memory to the screen. It can also be used to set breakpoints, single step and run through the code until a breakpoint is hit.
To run the debugger either a hex file or a source file must be given:
# -T turns debugging mode on
./h2 -T -r file.hex # Run simulator
Both modes of operation can be augmented with a symbols file, which lists where variables, labels and functions are located with the assembled core.
When the "-T" option is given debug mode will be entered before the simulation is executed. A prompt should appear and the command line should look like this:
$ ./h2 -T -R h2.fth
Debugger running, type 'h' for a list of command
debug>
Break points can be set either symbolically or by program location, the 'b' command is used to set breakpoints:
Numbers can be entered in octal (prefix the number with '0'), hexadecimal (prefix with '0x') or in decimal. As an example, the following three debug commands all set a breakpoint at the same location:
debug> b 16
debug> b 0x10
debug> b 020
'k' can be used to list the current break points that are set:
debug> k
0x0010
This sets a breakpoint when the function "key?" is called:
debug> b key?
Functions and labels can both be halted on, this requires either a symbols file to be specified on the command line or assemble and run to be used on a source file, not a hex file. Symbol files can be used on source or on hex files.
To single step the 's' command can be given, although not much will happen if tracing is turned off (tracing is off by default). Tracing can be toggled on or off with the 't' command:
debug> s
debug> s
debug> t
trace on
debug> s
0001: pc(089a) inst(4889) sp(0) rp(0) tos(0000) r(0000) call 889 init
debug> s
0002: pc(0889) inst(807a) sp(0) rp(1) tos(0000) r(089b) 7a
debug> s
0003: pc(088a) inst(e004) sp(1) rp(1) tos(007a) r(089b) 6004
It is advisable to turn tracing off when running issuing the 'c', or continue, command.
The '.' command can be used to display the H2 cores internal state:
debug> .
Return Stack:
0000: 0000 08aa 0883 017b 0000 031b 0000 ffb0 0000 02eb ffb5 0210 0167 0167
0167 0167
0010: 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
0000 0000
Variable Stack:
tos: 0000
0001: 0000 0000 0000 0001 0004 0005 0000 ffb0 0000 0000 0000 0000 0000 0000
0000 0000
0011: 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
0000 0000
pc: 0538
rp: 0001
dp: 0000
ie: false
And the 'p' command can be used to display the state of the simulated peripherals:
debug> p
LEDS: 00
VGA Cursor: 0005
VGA Control: 007a
Timer Control: 8032
Timer: 001b
IRC Mask: 0000
UART Input: 6c
LED 7seg: 0005
Switches: 00
LFSR: 40ba
Waiting: false
For a complete list of commands, use the 'h' command.
Other ways to enter debug mode include putting the ".break" assembler directive into the source code (this only works if the assemble and run command is used on source files, not on hex files), and hitting the escape character when the simulator is trying to read data via the simulated UART or PS/2 keyboard (the escape will still be passed onto the simulator, but it also activates debug mode).
A separate program can be compiled, tested under Linux and Windows. This simulates the Nexys3 board peripherals that the SoC interfaces with, but provides a graphical environment, unlike the command line utility. It is easier to interact with the device and see what it is doing, but the debugging sessions are a less controlled. It requires free glut.
- VGA shown on screen.
- UART or PS/2 input (selectable by pressing F11) comes from typing in the screen, and in the case of the UART this is buffered with a FIFO.
- UART output gets written to a display box.
- There are four 7-Segment displays as on the original board.
- The switches and push buttons can take their input from either keyboard keys or from mouse clicks.
- The LED indicators above the switches can be lit up.
Below is an image of a running session in the GUI simulator:
Building can be done with
make gui
And running:
make gui-run
Or:
./gui h2.hex (on Linux)
gui.exe h2.hex (on Windows)
The Linux build should work when the development package for free glut is installed on your system, the Windows build may require changes to the build system and/or manual installation of the compiler, libraries and headers.
The current key map is:
Up Activate Up D-Pad Button, Release turns off
Down Activate Down D-Pad Button, Release turns off
Left Activate Left D-Pad Button, Release turns off
Right Activate Right D-Pad Button, Release turns off
F1 - F8 Toggle Switch On/Off, F1 is left most, F8 Right Most
F11 Toggle UART/PS2 Keyboard Input
F12 Toggle Debugging Information
Escape Quit simulator
All other keyboard keys are redirected to the UART or PS/2 Keyboard input.
The Switches and D-Pad buttons can be clicked on to turn them on, the switches turn on with left clicks and off with right clicks. The D-Pads buttons turn on with a click on top of them and turn off with a key release anywhere on the screen.
The VHDL components used in this system are designed to be reusable and portable across different toolchains and vendors. Hardware components, like block RAM, are inferred and not explicitly instantiated. The components are also made to be as generic as possible, with most having selectable widths. This would be taken to the extreme, but unfortunately many vendors still do not support the VHDL-2008 standard.
| File | License | Author | Description |
|---|---|---|---|
| util.vhd | MIT | Richard J Howe | A collection of generic components |
| h2.vhd | MIT | Richard J Howe | H2 Forth CPU Core |
| uart.vhd | MIT | Richard J Howe | UART TX/RX (Run time customizable) |
| vga.vhd | LGPL 3.0 | Javier V García | Text Mode VGA 80x40 Display |
| Richard J Howe | (and VT100 terminal emulator) | ||
| kbd.vhd | ??? | Scott Larson | PS/2 Keyboard |
The pseudo Forth like language used as an assembler is described above, the application that actually runs on the Forth core is in itself a Forth interpreter. This section describes the Forth interpreter that runs on H2 Core, it is contained within embed.fth.
TODO:
- Describe the Forth environment running on the H2 CPU.
There are several languages used throughout this project, all of which are radically different from each other and require their own set of coding standards and style guides.
Common signal names:
clk - The system clock
rst - A reset signal for the module
we - Write Enable
re - Read Enable
di - Data In
din - Data In
do - Data Out
dout - Data Out
control - Generally an input to a register, the documentation
for the module will need to be consulted to find out
what each bit means
signal_we - The write enable for 'signal'
signal_i - This is an input signal
signal_o - This is an output signal
Generally the use of the "_i" and "_o" suffixes are not used, modules are kept short and names chosen so their meaning is obvious. This rule might be revisited once the project grows.
Components should:
- Be as generic as possible
- Use an asynchronous reset
- If a feature of a module can be made optional, by either ignoring outputs or setting inputs to sensible values, it should be.
- Where possible use a function, it is easy enough to turn a generic component into a module that can be synthesized but not the other way around.
- Use "downto" not "to" when specify variable ranges.
- Use assertions throughout the code with the correct severity level ('failure' for when something has seriously gone wrong or 'error' for debugging purposes)
- Constrain types and generic parameters if possible, as an example, if a generic value should never be zero, use "positive" not "natural".
- Try not to specify constants with fixed lengths where an expression using "others" can be used instead, for example:
constant N: positive := 4;
signal a: std_logic_vector(N - 1 downto 0) := (others => '1');
Instead of:
signal a: std_logic_vector(3 downto 0) := x"F";
The style rules are as follows:
- All words, including keywords, are to be in lower case. An underscore will separate words in names.
- Tabs are to be used to indent text, a tab spacing of 8 has been used when making the VHDL code
- Do not repeat the name of a entity, component, function or architecture, there is little point of repeating this, it just means when a unit has to be renamed it has to be done in two places instead of one.
- The ":" in definitions of signals belongs next to the signal name, not some arbitrary amount of spaces after it.
- Group related signals.
- Try to line up rows of signals
- Trigger logic on the rising edge, and use the "rising_edge" function not "clk'event and clk ='1'"
- By and large, each warning produced by the synthesis tool should be justified, and there should be very few warnings in the entire project if any.
- Do not use inferred latches.
- Load data from a file instead of generating VHDL files that contain the data, synthesis tools can handle impure VHDL functions that can read the initial data (for a ROM or block RAM as an example) from textual files.
An example of the formatting guidelines, this describes a simple arbitrary width register:
-- Lots of comments about what the unit does should go
-- here. Describe the waveforms, states and use ASCII
-- art where possible.
library ieee, work;
use ieee.std_logic_1164.all;
use ieee.numeric_std.all; -- numeric_std not std_logic_arith
entity reg is -- generic and port indented one tab, their parameters two
generic (
N: positive); -- Generic parameters make for a generic component
port (
clk: in std_logic; -- standard signal names
rst: in std_logic; --
we: in std_logic;
di: in std_logic_vector(N - 1 downto 0);
do: out std_logic_vector(N - 1 downto 0)); -- note the position of ");
end entity; -- "end entity", not "end reg"
architecture rtl of reg is
signal r_c, r_n: std_logic_vector(N - 1 downto 0) := (others => '0');
begin
do <= r_c;
process(rst, clk)
begin
if rst = '1' then -- asynchronous reset
r_c <= (others => '0');
elsif rising_edge(clk) then -- rising edge, not "clk'event and clk = '1'"
r_c <= r_n;
end if;
end process;
process(r_c, di, we)
begin
r_n <= r_c;
if we = '1' then
r_n <= di;
end if;
end process;
end; -- "end" or "end architecture"
There is quite a lot of C code used within this project, used to make a tool chain for the H2 core and to simulate the system.
- Usage of assertions for any pre or post condition, or invariant, are encouraged.
- Tabs are to be used instead of spaces, a tab width of 8 was used when coding the C, if this causes any code to go off screen then there is a problem with the code and not the tab length.
- Generally the K&R style is followed.
- Line lengths should ideally be limited to 80 characters, but this is definitely not an enforced limit.
- Where there are two or more data structures that must be kept in sync, with a one to one correspondence of elements, such as an enumeration and an array of strings that each enumeration maps onto, an X-Macro should be used to keep the data in sync and to initialize the enumeration and array of strings.
- Try to use only portable constructs and isolate the constructs that are not portable.
There is nothing too surprising about the C code within here, so some of the exceptions should be dealt with.
- Switch statements are formatted depending upon what the switch statement 'case' clauses look like, if they are a simple one liner such as an assignment or a mapping then the entire statement should occupy only a single line, for example:
static const char *alu_op_to_string(uint16_t instruction) {
/* notice also that the 'case' clauses are inline with the
* switch selector */
switch (ALU_OP(instruction)) {
case ALU_OP_T: return "T";
case ALU_OP_N: return "N";
case ALU_OP_T_PLUS_N: return "T+N";
case ALU_OP_T_AND_N: return "T&N";
case ALU_OP_T_OR_N: return "T|N";
case ALU_OP_T_XOR_N: return "T^N";
case ALU_OP_T_INVERT: return "~T";
case ALU_OP_T_EQUAL_N: return "N=T";
case ALU_OP_N_LESS_T: return "T>N";
case ALU_OP_N_RSHIFT_T: return "N>>T";
case ALU_OP_T_DECREMENT: return "T-1";
case ALU_OP_R: return "R";
case ALU_OP_T_LOAD: return "[T]";
case ALU_OP_N_LSHIFT_T: return "N<<T";
case ALU_OP_DEPTH: return "depth";
case ALU_OP_N_ULESS_T: return "Tu>N";
case ALU_OP_ENABLE_INTERRUPTS: return "seti";
case ALU_OP_INTERRUPTS_ENABLED: return "iset?";
case ALU_OP_RDEPTH: return "rdepth";
case ALU_OP_T_EQUAL_0: return "0=";
case ALU_OP_CPU_ID: return "cpu-id";
default: return "unknown";
}
}
- Unnecessary braces are avoided:
if (foo)
bar();
else
baz();
- "goto" can be used - it can be misused, but using it does not instantly make code inscrutable contrary to popular belief.
- Even better than using the embed project directly, would be to port the embed project so the meta-compiler runs directly on the hardware. The simulator could then be used to assemble new images, making the system (much more) self-hosting. Input/Output would be a problem, a possible solution is to use one of the UARTs for reading the meta-compiler and meta-compiled eForth program, and writing status/error messages. A second UART could be used to dump the binary as a stream of hexadecimal numbers, the simulator could redirect the second UART output to a file.
- Create a cut down version of the project; remove nearly everything apart from the H2 Core, Block RAM and timer components. The interrupt handler could be simplified as well. The UART could be handed in the H2 Core
- The GUI simulator could be written to be built against SDL, and include proper textures for the buttons and displays, instead of the current simulator which looks like an early 90s test application for OpenGL.
- Prepare more documentation. Specifically about the eForth interpreter that runs on the target and the online help stored within the non-volatile storage on the board.
- An IDE for resetting/uploading the image to the target board and then sending a text buffer to it would help in developing code for the platform.
- A Super Optimizer could be made for the H2.
- More instructions can be combined
- It might be possible to add a conditional exit instruction. Other instructions which would be useful are: Add with Carry, Bit Count, Leading Zeroes Count, Sign Extend, Arithmetic Right Shift, Rotate Left/Right, ...
- Add notes about picocom, and setting up the hardware:
picocom --omap delbs -b 115200 -e b /dev/ttyUSB1
- https://nanode0000.wordpress.com/2017/04/08/exploring-the-j1-instruction-set-and-architecture/
- https://www.fpgarelated.com/showarticle/790.php
- https://opencores.org/
- https://en.wikipedia.org/wiki/Peephole_optimization
- https://en.wikipedia.org/wiki/Superoptimization
- https://github.com/samawati/j1eforth
- https://github.com/jamesbowman/j1
- http://sovietov.com/app/forthwiz.html