The Ngaro Virtual Machine

Abstract

Ngaro is a minimalistic virtual machine emulating a dual stack computer with a simple instruction set and a few basic I/O devices.

At present, there are full implementations in Assembly (x86), C, F#, Go, Common Lisp, PHP, Python, and Ruby, and minimal implementations in C#, Forth, Lua, JavaScript, Java, Perl, and Scheme.

Memory Access

Ngaro provides a memory space consisting of 32-bit, signed integer values. The first address is mapped to zero, with subsequent values following in a strictly linear fashion.

Each addressable 32-bit unit is called a cell. There is no support in the instruction set for accessing values larger or smaller than a single cell.

Registers

Ngaro exposes no registers directly.

Stacks

Ngaro emulates two LIFO stacks. The primary one is the data stack, and serves as a place for holding data, and passing data between instructions.

The other is the address stack, which is used to hold return addresses from calls, and can be used to temporarily hold small amounts of data.

The current depth of a stack can be determined by using the I/O ports to query the VM.

A standard implementation will provide at least 128 cells for data stack, and 1024 for the address stack.

Image Files

On startup, Ngaro will load an image file. This is a flat, linear array of signed integer values. On disk, images are stored in little endian format, but the VM may convert them to big endian internally.

If an implementation uses big endian internally, this must be reported via the queries I/O port.

The first value loaded is mapped to address zero, and subsequent values are loaded to sequential addresses in the Ngaro memory space.

An image file does not contain copies of the stacks or any internal registers used by the VM.

I/O Ports

Interfacing with the underlying OS and hardware devices is done by reading and writing I/O ports.

Ports 0 - 12 are reserved for current and future standard I/O devices. A VM may provide non-standard devices on port numbers above or below this range.

Port Used For
0 Triggering I/O events
1 Keyboard
2 Text Output
3 Force Video Update
4 File I/O
5 Querying the VM
6 Canvas
7 Mouse

Port 0: Wait for Hardware Event

This port is used to synchronize communication between the ngaro and the virtual hardware devices.

See the description of the WAIT opcode for details.

Usage Example:

#0 #0 out, wait,

Port 1: Read from the Keyboard

To read a value from the keyboard, set port 1 to 1, wait for an I/O event, then read the key value from port 1.

Usage Example:

#1 #1 out,
#0 #0 out, wait,
#1 in,

Port 2: Character Generator

Ngaro provides a hardware character generator. This takes data off the stack, so you should make sure the character value is on the stack before using it. To use this, write the value 1 to port 2 and wait for an I/O event.

When a negative value is passed, the VM should clear the screen and position the cursor in the upper left corner.

Usage Example:

#98 #1 #2 out,
#0 #0 out, wait,

Port 3: Force Video Update

To help improve performance, an Ngaro implementation is permitted to cache output and update the display periodically. For implementations that do this, seting this port will force an immediate update.

An implementation that caches output must check this routinely, and respond immediately. No waiting for an I/O event is neccessary for this.

Example:

#0 #3 out,

Port 4: File Operations

If your Ngaro implementation allows saving images, you can use this port to do so. To save, set port 4 to 1 and wait.

Other operations can be done using negative values, if the VM supports this. To use these, setup the stack, write the operation code to port 4, wait, then read the value back in from port 4.

Op Takes Returns Notes
1
0 Save the image
2 filename 0 Include a file ( like --with )
-1 filename, mode handle Open a file
-2 handle byte Read a byte from a file
-3 character, handle flag Write a byte to a file
-4 handle flag Close a file
-5 handle offset Return current location in file
-6 offset, handle flag Seek a new location in file
-7 handle size Return the size of a file
-8 filename flag Delete a file.

Valid modes for opening files are:

Value Used For Create if not existing?
0 Open file for reading No
1 Open file for writing Yes
2 Open file for append Yes
3 Open file for modification No

Reading and modification should not create a file if none exist. Writing and append modes should create a file if it does not exist.

The write mode should create a new file, removing the contents of existing files with the same name.

The append mode should set the file read/write position to the end of the file.

If opening fails, the returned handle should be zero. Any non-zero handle is considered valid.

When closing a valid handle, close should return zero.

The write operation should return a value of 1. Any other value indicates an error.

The delete operation should return -1 if the file is deleted, or 0 if the deletion fails.

Port 5: Queries Into the VM Devices

Set port 5 to one of the following values; wait; then read the result back.

value returns
-1 Memory Size
-2 Does a Canvas device exist?
-3 Canvas Width
-4 Canvas Height
-5 Data Stack Depth
-6 Address Stack Depth
-7 Does a Mouse device exist?
-8 Current time (in seconds, Unix-style)
-9 Exit the VM
-10 Query for an enivronment variable
-11 Console Width
-12 Console Height
-13 Number of bits per cell
-14 0 for little endian, 1 for big endian
-15 -1 if Port 8 enabled, 0 if disabled
-16 Return max depth of data stack
-17 Return max depth of address stack

At a minimum, an implementation must provide support for -1, -5, -6, -8, and -9.

For -5 and -6, "depth" refers to the number of items on the specified stack.

For -10, the application must provide a buffer address on the stack, and a pointer to a string. The VM should search the system environment for the string and copy its value to the application memory, starting at the provided buffer address. If an environment variable is not found, the VM should store a value of zero in the provided buffer address.

For -13, if the returned value is zero, the image can assume a 32-bit environment.

For -14, if the VM is using big endian internally, this should return a value of 1.

Port 6: Canvas

Some Ngaro implementations allow for drawing to a canvas device. Setup the data stack as shown in the table, and write the appropriate values to port 6.

This device is optional.

value stack action performed
1 n- set color for drawing operations
2 xy- draw a pixel at coordinates x, y
3 xyhw- draw a rectangle of specified width (w) and height (h). The top corner is denoted by the x, y pair
4 xyhw- draw a filled rectangle of specified width (w) and height (h). The top corner is denoted by the x, y pair
5 xyh- draw a vertical line of height (h) starting at x, y
6 xyw- draw a horizontal line of width (w) starting at x, y
7 xyw- draw a circle of width (w) starting at x, y
8 xyw- draw a filled circle of width (w) starting at x, y

For setting colors, the following values are guarateed safe:

code name
0 black
1 dark blue
2 dark green
3 dark cyan
4 dark red
5 purple
6 brown
7 dark gray
8 gray
9 blue
10 green
11 cyan
12 red
13 magenta
14 yellow
15 white

Additional colors may be supported, but are not guaranteed to exist.

Port 7: Mouse

Set port 7 to one of the following values and wait. The results will be pushed to the data stack.

This device is optional.

value returns
1 Mouse X and Y coordinates. Y will be on TOS when done. X will be NOS.
2 Is mouse button pressed? 0 = false, non-zero is true. True values may indicate the button being pressed, but this is not required.

Port 8: Enhanced Text Console

Set port 8 to one of the following values and wait.

This device is optional.

Due to platform constraints, implementations may not offer all of these functions. An implementation providing this port should at least support cursor positioning. Any non-implemented functions should be silently ignored.

An implementation may provide additional, non-standard functionality using negative values.

value stack action
1 rc- Move the cursor to the specified row and column
2 n- Set the foreground text color
3 n- Set the background color

The VM is expected to understand the following colors:

value color
0 Black
1 Red
2 Green
3 Yellow
4 Blue
5 Magenta
6 Cyan
7 White

Instruction Set

One instruction per memory location. Instructions with an x in the A column take an additional value in the following memory location.

All opcode numbers are listed in decimal. Stack diagrams are for the data stack.

opcode name assembler A stack
0 NOP nop,   -
1 LIT lit, x -n
2 DUP dup,   n-nn
3 DROP drop,   n-
4 SWAP swap,   xy-yx
5 PUSH push,   n-
6 POP pop,   -n
7 LOOP loop, x n-n
8 JUMP jump, x -
9 RETURN ;,   -
10 GT_JUMP >jump, x xy-
11 LT_JUMP <jump, x xy-
12 NE_JUMP !jump, x xy-
13 EQ_JUMP =jump, x xy-
14 FETCH @,   a-n
15 STORE !,   na-
16 ADD +,   xy-z
17 SUBTRACT -,   xy-z
18 MULTIPLY *,   xy-z
19 DIVMOD /mod,   xy-rq
20 AND and,   xy-z
21 OR or,   xy-z
22 XOR xor,   xy-z
23 SHL <<,   xy-z
24 SHR >>,   xy-z
25 ZERO_EXIT 0;   n-?
26 INC 1+,   x-y
27 DEC 1-,   x-y
28 IN in,   p-n
29 OUT out,   np-
30 WAIT wait,   -

Instruction Processing

The instruction pointer is incremented, then the opcode at the current address is handled. Execution ends when the instruction pointer is greater than the end of the simulated memory space.

A psuedocode in Retro:

-1 !ip
[ ip ++ processOpcode @ip 1000000 < ] while

And in Lua:

ip = 0
while ip < 1000000 do
  processOpcode()
  ip = ip + 1
end

The Instructions

Opcode 0: NOP

Does nothing.

Opcode 1: LIT

Push the value in the following memory location to the data stack. Advances the instruction pointer by one.

In memory this might appear as:

0000 LIT
0001 101

After LIT executes, the IP would be set at 0001, and the top item on the data stack would be 101.

Opcode 2: DUP

Make a duplicate copy of the top item on the data stack and push the copy to the data stack.

before after
1
2
3
1
1
2
3

Opcode 3: DROP

Remove the top item from the data stack.

before after
1
2
3
2
3

Opcode 4: SWAP

Remove the top two items from the stack, and push them back in the reverse order.

before after
1
2
3
2
1
3

Opcode 5: PUSH

Remove the top item from the data stack, and push it to the address stack.

before after
1
2
3
2
3

Opcode 6: POP

Remove the top item from the address stack, and push it to the data stack.

before after
2
3
1
2
3

Opcode 7: LOOP

Decrement the top value on the stack and advance the instruction pointer. If the top item on the stack is greater than zero, jump to the address following the LOOP instruction, otherwise discard the top item on the stack and continue execution normally.

Opcode 8: JUMP

Set the instruction pointer to the address in the cell following the JUMP instruction.

This instruction needs to decrement the requested address by one to account for the increment of the instruction pointer by the opcode process cycle. E.g., if the jump target is 1234, JUMP needs to set the instruction pointer to 1233.

To improve performance, this instruction may skip leading NOP's at the destination address.

Opcode 9: RETURN

Return from a call to a subroutine. This will pop the return address from the address stack, and set the instruction pointer to it.

Opcode 10: GT_JUMP

Increment the instruction pointer.

Pop the top two values from the stack. If the first stack item is greater than the second item, set the instruction pointer to the address stored at the memory location following this instruction. If not, continue execution.

This instruction needs to decrement the requested address by one to account for the increment of the instruction pointer by the opcode process cycle. E.g., if the jump target is 1234, the instruction pointer should be set to 1233.

In memory, this will be stored as:

0000 GT_JUMP
0001 destination
before after
1
2
 

Opcode 11: LT_JUMP

Increment the instruction pointer.

Pop the top two values from the stack. If the first stack item is less than the second item, set the instruction pointer to the address stored at the memory location following this instruction. If not, continue execution.

This instruction needs to decrement the requested address by one to account for the increment of the instruction pointer by the opcode process cycle. E.g., if the jump target is 1234, the instruction pointer should be set to 1233.

In memory, this will be stored as:

0000 LT_JUMP
0001 destination
before after
1
2
 

Opcode 12: NE_JUMP

Increment the instruction pointer.

Pop the top two values from the stack. If the first stack item is not equal to the second item, set the instruction pointer to the address stored at the memory location following this instruction. If not, continue execution.

This instruction needs to decrement the requested address by one to account for the increment of the instruction pointer by the opcode process cycle. E.g., if the jump target is 1234, the instruction pointer should be set to 1233.

In memory, this will be stored as:

0000 NE_JUMP
0001 destination
before after
1
2
 

Opcode 13: EQ_JUMP

Increment the instruction pointer.

Pop the top two values from the stack. If the first stack item is equal to the second item, set the instruction pointer to the address stored at the memory location following this instruction. If not, continue execution.

This instruction needs to decrement the requested address by one to account for the increment of the instruction pointer by the opcode process cycle. E.g., if the jump target is 1234, the instruction pointer should be set to 1233.

In memory, this will be stored as:

0000 EQ_JUMP
0001 destination
before after
1
2
 

Opcode 14: FETCH

Remove the top item from the data stack. Lookup the value stored in the memory address this value points to, and push the value read to the data stack.

Assuming that memory at 1234 contains 45:

before after
1234
45

Opcode 15: STORE

Take two values from the stack. The top item will be a destination address, and the second will be a value. Modify the contents of the specified memory address to be equal to the value.

before after
1
2
 

In this, 1 would be the address, and 2 would be the value to store there.

Opcode 16: ADD

Take two values from the data stack, add them together, and push the results to the data stack.

before after
1
2
3

Opcode 17: SUBTRACT

Take two values from the data stack. Subtract the top item from the second item, and push the results back to the data stack.

before after
4
9
5

Opcode 18: MULTIPLY

Take two values from the data stack. Multiply them and push the results back to the data stack.

before after
2
3
6

Opcode 19: DIVMOD

Take two values from the data stack. The top item is the divisor, and the second item is the dividend. Perform the division, and push the quotient and remainder to the stack. After execution the quotient should be on top, with the remainder below it.

Division is symmetric, not floored.

before after
2
5
2
1

Opcode 20: AND

Remove the top two items on the data stack. Perform a bitwise AND operation and push the result back to the data stack.

before after
-1
-1
-1
before after
0
-1
0
before after
0
0
0

Opcode 21: OR

Remove the top two items on the data stack. Perform a bitwise OR operation and push the result back to the data stack.

before after
-1
-1
-1
before after
0
-1
-1
before after
0
0
0

Opcode 22: XOR

Remove the top two items on the data stack. Perform a bitwise XOR operation and push the result back to the data stack.

before after
-1
-1
0
before after
0
-1
-1
before after
0
0
0

Opcode 23: SHL

Take two values from the data stack. Perform a bitwise left shift on the second item by the number of bits specified by the top item. Push the results back to the data stack.

The values in these tables are in binary.

before after
11
111000111
111000111000

The results of a negative shift are implementation specific.

Opcode 24: SHR

Take two values from the data stack. Perform a bitwise right shift on the second item by the number of bits specified by the top item. Push the results back to the data stack.

The values in these tables are in binary.

before after
11
111000111000
111000111

The results of a negative shift are implementation specific.

Opcode 25: ZERO_EXIT

If the top item on the stack is zero, remove and discard it, then pop the top item from the address stack and set the instruction pointer to it.

If the top item is not zero, leave it alone and do nothing.

before after
1
1
before after
0
 

Opcode 26: INC

Increase the value on the top of the stack by one.

before after
2
3

Opcode 27: DEC

Decrease the value on the top of the stack by one.

before after
2
1

Opcode 28: IN

Take a value from the data stack. Read the value stored in the I/O port corresponding to the value read from the stack, and push this value to the data stack.

After reading, set the value of the port read to zero.

before after
1
?

The value returned will vary depending on the I/O device subsystem and specific port requested.

Opcode 29: OUT

Take two values off the data stack. The top value will be an I/O port number, and the second will be a value. Store the value in the I/O port specified.

before after
2
1
 

With the values in this table, port 2 would be set to a value of 1.

Opcode 30: WAIT

Run the simulated device handler, provided that port 0 is set to 0, and at least one other port is set to a value other than 0. Otherwise, continue execution as normal.

If the device handler runs, it will setting port 0 to 1, thus preventing itself from running again until the code sets port 0 back to 0. Therefore, programs should always set I/O port 0 to 0 before invoking WAIT.

Opcodes Above 30

Any opcode above 30 is treated as an implicit call.

The VM will push the current value of the instruction pointer to the address stack and set the instruction pointer to the value of the opcode. Note that like jumps, the VM must decrement this value by one to account for the increment that will happen before the next instruction is processed.