Whilst reading But How Do It Know? by J. Clark Scott I felt compelled to write something to simulate the computer the book describes.
Starting from NAND gates, and moving up through to registers, RAM, the ALU, the control unit and adding I/O support, I eventually ended up with a fully functional machine.
All the components of the system are based on logic gates and the way they are connected together via the system bus.
For a write up about this project, see my blog post about it here https://djharper.dev/post/2019/05/21/i-dont-know-how-cpus-work-so-i-simulated-one-in-code/
~0.006mhz
- at least on my machine
- 16-bit
- the book describes an 8-bit CPU for simplicity but I wanted more RAM and there is only one system bus
- 65K RAM
- 240x160 screen resolution
- 4x 16-bit registers (
R0
,R1
,R2
,R3
)
Missing features
- Interrupts, so you have to write awful polling code
- The book does shortly describe how to extend the system to support interrupts but would involve a lot more wiring
- Stack pointer register + stack + stack manipulation instructions so nested
CALL
instructions won't work and registers may be left in an inconsistent state - Hard drive
- Subtract instruction
MOV
instruction- Floating point math (lol)
- Everything else you could think of from a modern CPU
Bonus features
- No Meltdown/SPECTRE risk
- Can easily overwrite any portion of memory without any protective mode getting in the way
- Currently incapable of accessing the internet
- I can see how you might write a simple networking I/O adapter, although I'd imagine it would be tedious writing the assembly to get bytes in and out of it 🤔
Instruction | Type | Description | Example |
---|---|---|---|
LOAD Ra, Rb |
Machine | Load value of memory address in register A into register B | LOAD R1, R2 |
STORE Ra, Rb |
Machine | Store value of register B into memory address in register A | STORE R3, R1 |
DATA Ra, <VALUE> |
Machine | Put <VALUE> into register A. <VALUE> can either be a symbol, prefixed with % (e.g. %LINE-X ) or a numeric value (e.g. 0x00F2 or 23 ) |
DATA R3, %KEYCODE |
JR Ra |
Machine | Jump to instruction in memory address in register A | JR R2 |
JMP <LABEL> |
Machine | Jump to instruction in memory address for <LABEL> |
JMP startloop |
JMP[CAEZ]+ <LABEL> |
Machine | Jump to instruction in memory address for <LABEL> if flags register for any combination of CAEZ is true |
JMPEZ endloop |
CLF |
Machine | Clear contents of flags register | CLF |
IN <MODE>, Ra |
Machine | Request input from IO device to Register A | IN Data, R3 |
OUT <MODE>, Ra |
Machine | Send output to IO device for register A | OUT Addr, R2 |
ADD Ra, Rb |
Machine | 16 bit addition of two registers | ADD R0, R2 |
SHR Ra |
Machine | Shift right register A | SHR R0 |
SHL Ra |
Machine | Shift left register A | SHL R0 |
NOT Ra |
Machine | Bitwise NOT on register A | NOT R2 |
AND Ra, Rb |
Machine | Bitwise AND on two registers | AND R2, R3 |
OR Ra, Rb |
Machine | Bitwise OR on two registers | OR R0, R1 |
XOR Ra, Rb |
Machine | Bitwise XOR on two registers | XOR R1, R0 |
CMP Ra, Rb |
Machine | Compare register A and register B (will set flags register) | CMP R1, R2 |
CALL <LABEL> |
Pseudo | Call a subroutine. This will jump to the subroutine, on completion, the subroutine should jump back and continue from the next instruction. Note: there is no stack functionality here so all registers may be in a different state at the end of the subroutine. | CALL pollKeyboard |
The following I/O devices are supported by the computer.
Device | Address |
---|---|
Keyboard | 0x000F |
Display | 0x0007 |
There is no memory management unit or protected areas of memory.
However the assembler and simulator will start executing user code from offset 0x0500
Machine code can be written in text and assembled using a crude assembler I wrote.
See assembler for more information.
@realkompot made an awesome compiler https://github.com/realkompot/llvm-project-scott-cpu for this using LLVM that produces working binaries to run on the simulator, check out the cool little snake game example https://github.com/realkompot/llvm-project-scott-cpu/tree/scott-cpu/_scott-cpu
Requirements
- go 1.12+
- GLFW 3.2+
Building:
make
There are some unit tests that take 30-45 seconds to run through, by running
make test
The computer can be run using the wrapper tool I wrote that utilises GLFW for I/O functionality.
Example of running the brush.bin
program
./bin/simulator -bin _programs/brush.bin
You can see some example programs I wrote under _programs/, note the ASM code I wrote for these is very bad and I lost my sanity a bit when writing them.
I'm taking myself on a journey, a hardware journey you might say. I want to understand how computers work at a lower level but not quite low enough for the physics/digital electronics side of things.
Just enough to see all the pieces of the system interacting. I remember doing a lot of this stuff in school but I'd say my education seemed to focus on the concepts (Von-Neumann architecture, fetch-decode-execute) rather than the actual construction of a CPU.
This simple computer is the start of that journey, it's actually been a very rewarding little project.
I hope to move onto playing around with X86/ARM/RISC-V next although I suspect it will be quite a leap (of faith)