|
|
|
|
# Mistakes
|
|
|
|
|
|
|
|
|
|
## Don't design opcodes in your head
|
|
|
|
|
|
|
|
|
|
You WILL screw up and your binary encodings WILL overlap.
|
|
|
|
|
You definitely need some tool to design instruction opcodes
|
|
|
|
|
for anything more complicated than just 1-byte instructions.
|
|
|
|
|
|
|
|
|
|
Turned out this mistake propogates further when implementing
|
|
|
|
|
assembler.
|
|
|
|
|
|
|
|
|
|
## Load/store instructions MUST be able to use register as address
|
|
|
|
|
|
|
|
|
|
I have no idea how we didn't notice this mistake.
|
|
|
|
|
I guess it is not too bad for a CPU with 16 bytes of memory,
|
|
|
|
|
but for something bigger it is unacceptable.
|
|
|
|
|
|
|
|
|
|
## Jump instructions need more love
|
|
|
|
|
|
|
|
|
|
Jumping 3 instructions back and 7 forward is too limiting.
|
|
|
|
|
Seriously, that's stupid.
|
|
|
|
|
|
|
|
|
|
Also, in my current design, jumping offset of 0 will
|
|
|
|
|
move CPU to an infinite loop.
|
|
|
|
|
|
|
|
|
|
## Implementing assembler with regex is not fancy
|
|
|
|
|
|
|
|
|
|
It is probably easier to implement than writing proper
|
|
|
|
|
lexer/parser, but error reporting can be weird.
|
|
|
|
|
|
|
|
|
|
## Two registers are really not enough
|
|
|
|
|
|
|
|
|
|
Just imagine, you are trying to write a loop, you use
|
|
|
|
|
`R1` as your index variable and `R0` for your loop condition.
|
|
|
|
|
Now you have already ran out of registers and you didn't even
|
|
|
|
|
start writing loop body!
|
