187 lines
5.9 KiB
Org Mode
187 lines
5.9 KiB
Org Mode
#TITLE Project Specification
|
||
|
||
* Binary interface
|
||
|
||
The VM does not use floating point numbers, it instead uses fixed point numbers.
|
||
|
||
This is for portability reasons as some devices might not have a FPU in them
|
||
|
||
especially microcontrollers and some retro game systems like the PS1.
|
||
|
||
** Numbers
|
||
| type | size (bytes) | description |
|
||
|------+--------------+---------------------------------------|
|
||
| u8 | 1 | unsigned 8bit, alias =char= and =byte= |
|
||
| bool | 1 | unsigned 8bit, =false= or =true= |
|
||
| i8 | 1 | signed 8bit for interop |
|
||
| u16 | 2 | unsigned 16bit for interop |
|
||
| i16 | 2 | signed 16bit for interop |
|
||
| u32 | 4 | unsigned 32bit, alias =nat= |
|
||
| i32 | 4 | signed 32bit, alias =int= |
|
||
| f32 | 4 | signed 32bit fixed number, alias =real= |
|
||
|
||
* Memory
|
||
|
||
Uses a harvard style archecture, meaning the code and ram memory
|
||
are split up into two seperate blocks.
|
||
|
||
In the C version you can see these are two seperate arrays 'code' and 'mem'.
|
||
|
||
During compilation constants and local variables are put onto 'mem'
|
||
|
||
* Opcodes
|
||
|
||
| 2^n | count |
|
||
|----+-------|
|
||
| 2^3 | 8 |
|
||
| 2^4 | 16 |
|
||
| 2^5 | 32 |
|
||
| 2^6 | 64 |
|
||
| 2^8 | 128 |
|
||
|
||
** could be encoded
|
||
- op type [this is to maximize jump immidate and load immidate size]
|
||
- memory location
|
||
- local value / register
|
||
- local value type
|
||
|
||
*** Simplest
|
||
[opcode][dest][src1][src2]
|
||
[8][8][8][8]
|
||
|
||
*** Maximize inline jump and load-immidate
|
||
[0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0]
|
||
|
||
[0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 | 0 0 ] noop
|
||
[0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 1 | 0 0 ] call
|
||
[0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 1 0 | 0 0 ] return
|
||
[0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 1 1 | 0 0 ] syscall?
|
||
[0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 1][0 0 0 | 1 0 0 | 0 0 ] exit
|
||
|
||
[0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0][0 0 0 0 0 0 | 0 1 ] jump ~1GB range
|
||
[0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0][0 0 0 0 0 0 | 1 0 ] load-immidate 2^30 max
|
||
|
||
*** multibyte ops
|
||
- ones that would be easier if they were multibyte
|
||
- jump
|
||
- load immidate
|
||
- syscall
|
||
- call
|
||
|
||
|
||
0 0 - system, lowest because lower opcodes are faster
|
||
0 1 - memory
|
||
1 0 - math
|
||
1 1 - jump
|
||
|
||
[0 0][0 0 0 0 0 0] = [system][no op]
|
||
[0 0][1 0 0 0 0 0] = [system][loadimm]
|
||
[0 0][0 0 0 0 0 0] = [system][return]
|
||
|
||
|
||
J [0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0][0 0 0 0 0 0 | 0 1]
|
||
L [0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0][0 0 0 0 0 0 0 0][0 | o o o o o | 1 0]
|
||
|
||
R [r r r r r r | r r][r r r r | r r r r][r r | b b | 0 0 0 0][o o o o o o | 1 1]
|
||
|
||
|
||
|
||
[0 0 0 0 0 | r r r][r r r r | r r r r][r r | r r r r r r][b b | t | o o o | 1 1]
|
||
|
||
[0 0 0 | r r r r r][0 0 0 | r r r r r][0 0 0 | r r r r r][b b | t | o o o | 1 1]
|
||
[0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 | 1 | 0 0 1 | 1 1] [math][add][f][8]
|
||
[0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 | 1 | 0 1 0 | 1 1] [math][sub][f][8]
|
||
[0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 | 1 | 0 1 1 | 1 1] [math][mul][f][8]
|
||
[0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 | 1 | 1 0 0 | 1 1] [math][div][f][8]
|
||
[0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 0 | 0 0 0 0 0][0 0 | 1 | 1 0 1 | 1 1] [math][mod][f][8]
|
||
|
||
[0 1][0 0 1][0][0 0] = [math][sub][f][16]
|
||
[0 1][0 1 0][0][0 0] = [math][mul][f][32]
|
||
[0 1][0 1 1][0][0 0] = [math][div][f][?]
|
||
[0 1][1 0 0][0][0 0] = [math][and][f][?]
|
||
|
||
[1 1][0][0 0][0 0] = [jmp][u][eq]
|
||
[1 1][0][0 0][0 0] = [jmp][u][ne]
|
||
[1 1][0][0 0][0 0] = [jmp][u][lt]
|
||
[1 1][0][0 0][0 0] = [jmp][u][gt]
|
||
|
||
[1 1][1][0 0 0][0 0] = [jmp][s][le]
|
||
[1 1][1][0 0 0][0 0] = [jmp][s][ge]
|
||
[1 1][1][0 0 0][0 0] = [jmp][s][]
|
||
[1 1][1][0 0 0][0 0] = [jmp][s][]
|
||
|
||
3 2 2 1
|
||
[...] i 32 0
|
||
|
||
3 1 3 1
|
||
[...] u rl j
|
||
[...] u ab j
|
||
[...] u eq j
|
||
[...] u nq j
|
||
[...] u lt j
|
||
[...] u le j
|
||
[...] u gt j
|
||
[...] u ge j
|
||
|
||
[jmp][dest][18]
|
||
[lli][dest][2]
|
||
|
||
|
||
|
||
int 3
|
||
|
||
add
|
||
sub
|
||
mul
|
||
div
|
||
|
||
jeq
|
||
jne
|
||
|
||
jlt
|
||
jle
|
||
jgt
|
||
jge
|
||
|
||
Maybe opcodes could be Huffman encoded? That way the smaller opcodes get more operand data within the u32
|
||
|
||
At compile time each function gets N number of locals (up to 255). These are allocated onto memory along with everything else, but they come before the heap values.
|
||
|
||
Maybe instead of push pushing onto a stack it instead pushes onto a child frame? Pops would just mark what locals need to be replaced by the child function? Maybe we can get rid of call/return and just use jumps?
|
||
|
||
U8 u8 u8 u8
|
||
Push/pop parent_local child_local metadata
|
||
|
||
** Maybe more flexible calling convention?
|
||
|
||
Memory-to-memory with register characteristics?
|
||
|
||
Passed in values
|
||
|
||
Copy each argument from the callers local to the callees local. This includes pointers.
|
||
|
||
child modifies the heap
|
||
|
||
If a child modifies a value in the parents heap do nothing, this is expected behavior.
|
||
|
||
If a child changes the size of a parents heap then copy the heap value to the child’s frame.
|
||
|
||
Returned values
|
||
|
||
If a primitive value just copy from child local to parent local
|
||
|
||
If a heap value is returned but placed in a new local in the parent then copy the child to the parent and update the frames memory pointer
|
||
|
||
If a heap value is replaced (i.e. the return sets a heap value with its modified version) then
|
||
Sort each returned value by its pointers location in memory, lowest first
|
||
Move to position of returned values lowest ptr position.
|
||
Read fat ptr size of the earliest value.
|
||
Take the current size of heap.
|
||
Move to just after the end of the size + ptr.
|
||
Copy all values from that location through current end of heap to the old start location of that value.
|
||
Subtract the old size of the value from the mp.
|
||
Copy the new sized value and put it at the current end of the heap.
|
||
Update the new pointer’s local position.
|
||
Add the new size to the mp.
|
||
Repeat for each returned value that is a replaced heap value.
|