old-DasOS/documentation/spu-mark-1.md
2016-06-27 16:04:56 +02:00

14 KiB
Raw Blame History

SuperVM SPU Mark I

The SPU Mark I is a stack machine with a simple, but flexible command set. The abbreviation SPU stands for Stack Processing Unit, Mark I declares that it is the first version of the SPU.

Purpose of this document

This document is meant to give a complete overview over the concepts and abstract workings of the SPU Mark I.

It is targeted at uses who program the SPU Mark I with the native assembly language, system programmers who want to include the virtual machine in their system or create their own SPU Mark I implementation.

Concepts

SPU Mark I is a virtual CPU that emulates a 32 bit stack machine. Instead of utilizing registers operations take their operands from the stack and push their results to it.

An instruction is split into two parts: The instruction configuration and the command. The command defines what operation should be performed (memory access, calculation, ...), whereas the configuration defines the behaviour of instruction (stack/flag-modifications).

Memory Areas

The virtual machine has three separarated memory areas. Each area serves a specific purpose and should not overlap the others.

Code Memory

The code memory contains an immutable block of code that is instruction indexable. Each instruction is 64 bit wide.

Stack Memory

The virtual machine utilizes a stack to provide operands to instructions. This stack stores temporary values the program is working with.

Each entry on the stack is an 32 bit value that is mostly interpreted as a pointer, an index or an unsigned or signed integer. It is also possible to store a 32bit IEEE floating point number on the stack.

The size of the stack is defined by the implementation, but it should contain at least 1024 entries. This allows a fair recursive depth of 128 recursions with an average of 6 local variables per function call.

Data Memory

SPU Mark I also provides a memory model that allows storing RAM data that is accessible by different parts of the code. The data memory is byte accessible and can be written or read.

It is implementation defined how the memory is managed and accessible. It can be a sparse memory with different sections, it could utilize a software-implemented paging process or just be a flat chunk of memory.

Every pointer that accesses data memory (e.g. via store and load) contains the address of a byte in memory, starting with zero.

Registers and Flags

The SPU Mark I is a stack machine, but has also some control registers that can be set with special instructions. The registers mainly control stack access or control flow.

Each register has a size of 32 bits. Only exception is the flag register which contains a single bit per flag.

Mnemonic Register Function
SP Stack Pointer Stores the current 'top' position of the stack.
BP Base Pointer Stores the current stack frame position.
CP Code Pointer Stores the instruction which is executed next.
FG Flag Register Stores the state of the flags.

Stack, Base and Code Pointer store indexes instead of actual memory addresses. This prevents the VM to execute invalid instructions as the code pointer always points to the start of an instruction.

Unlike common on most of the current CPUs, the stack and base pointer are growing upwards, each push increments the stack pointer by one, each pop decrements it.

All registers start initialized with a zero.

Stack Pointer

The stack pointer points to the top of the stack. Each push operation increases the stack pointer by one, each pop operation reduces it by one.

Base Pointer and Function Calls

The base pointer is a pointer that can be set to access the stack relative to it. This relative access is done by the commands get and set.

The base pointer is designed to create stack frames for functions with local variables as it is not possible to access local variables on the stack with only push and pop operations.

Code Pointer

The code pointer contains the instruction which is executed next. Modifying the code pointer is equivalent to a jump operation.

Flag Register

Bit Flag Option
0 Zero Is set when the output is zero.
1 Negative Is set when the MSB is set.

Instructions

An SPU Mark I instruction is composed of multiple components:

Component Range Size Function
execution Z See below. 2 Excution dependend on Zero?
execution N See below. 2 Excution dependend on Negative?
input0 Zero/Pop/Peek/Arg 2 Where does input0 come from?
input1 Zero/Pop 1 Where does input1 come from?
command [6bit] 6 Which command is executed?
cmdinfo [16bit] 16 Parameter value for the command.
flagmod yes/no 1 Does this command modifies flags?
output Discard/Push/Jump 2 What is done with the output?
argument [32bit] 32 Some commands can take extra information

Execution Modes

The execution mode checks whether the instruction will be execution or not. The execution depends on the state of the flags. An X means "Don't care", a 0 means the flag must be cleared and a 1 means the flag must be set.

| State | Binary Representation | | X | 0b00 | | 0 | 0b10 | | 1 | 0b11 |

An instruction is only executed when all conditions are met.

Flag Range
Zero X/0/1
Negative X/0/1

Inputs

Each instruction has a defined set of inputs. The inputs are parameters for the executed commands and allow a flexible configuration.

The first input can utilize all input methods, the second only provides zero and pop.

# Method Description
0 Zero The input value is zero.
1 Pop The input value is popped from the stack.
2 Peek The input value is the top value of the stack.
3 Argument The instruction argument is copied to the input.

Commands

A command is the execution part of an instruction. It defines a core operation which does the effective calculations.

Each command can be seen as a function defined as:

output command(input0, input1, argument, cmdinfo)

The function has 4 inputs which can be used to calculate the output or change the vm state. Each command also has the option to output a specific value that can be processed further.

ID Command Action Description
0 COPY output = input0 Copies a value to the output.
1 STORE output = MEMORY[input0] = input1 Stores a value in process memory.
2 LOAD output = MEMORY[input0] Loads a value from process memory.
3 GET output = STACK[BP + input0] Reads a value from the stack with base pointer offset.
4 SET output = STACK[BP + input0] = input1 Writes a value to the stack with base pointer offset.
5 BPGET output = BP Gets the base pointer.
6 BPSET output = BP = input0 Sets the base pointer.
7 CPGET output = CP + cmdinfo Gets the current program counter with an offset.
8 MATH output = input0 OP[info] input1 Does an ALU operation.
9 SPGET output = SP + input0 Gets the current stack pointer.
10 SPSET output = SP + input0 = input1 Sets the current stack pointer.
11 SYSCALL output = SysCall(input0, input1) Calls an OS dependend operation.
12 HWIO output = HardwareIO(input0, input1) Calls an abstract hardware operation.

Copy

This command just copies the first input value to the output value. It can be used for a broad variety of instructions like modifying the stack, jumping or constant flag modification.

Store, Load

These command accesses the process memory. Store writes value to process memory, Load reads a value from it. The command info defines, what kind of value is written:

cmdinfo Value type Size in Bytes
0 uint8_t 1
1 uint16_t 2
2 uint32_t 4

Get, Set

Get and Set are used to modify the local stack frame. They allow modification of the stack around the base pointer with a given offset. Get reads a value from the stack, Set writes a value to the stack.

BpGet, BpSet

These commands modify the base pointer to set the stack offset for Get and Set.

CpGet

This command reads the current program counter and returns it. The program counter is also offsetted by the command info.

Math

The math command is a compound operator that contains all ALU operations. The ALU operation is selected by the cmdinfo.

If an ALU operation takes two operands, the right hand side is defined by input0, the left hand side is defined by input1. This allows a configuration such that the right hand side operand is taken by the argument of the instruction instead of beeing popped from the stack.

cmdinfo Operation Forumla
0 Addition input1 + input0
1 Subtraction input1 - input0
2 Multiplication input1 * input0
3 Division input1 / input0
4 Euclidean Division / Modulo input1 % input0
5 Bitwise Logic And input1 ∧ input0
6 Bitwise Logic Or input1 input0
7 Bitwise Logic Xor input1 ⊻ input0
8 Bitwise Logic Not ˜input0
9 Rotating Bit Shift Left input1 ⊲ input0
10 Rotating Bit Shift Right input1 ⊳ input0
11 Arithmetic Bit Shift Left input1 ≺ input0
12 Arithmetic Bit Shift Right input1 ≻ input0
13 Logic Bit Shift Left input1 « input0
14 Logic Bit Shift Right input1 » input0
15 Negation -input0

SpGet, SpSet

These commands modify the stack pointer directly. SpGet reads the stack pointer, SpSet writes the stack pointer.

SysCall

This command provides an interface to the executing host system. The effects, parameters and results for this command must be defined by the host.

HwIO

This command also provides an interface to the executiing host system, but is focused on hardware IO. The effects, parameters and results are also defined by the host.

Command Info

The cmdinfo part of the instruction is passed to the executing command adding some non-dynamic information for the execution.

Flag Modification

This part defines if the instruction should change the flags according to its output. The Zero flag is set when the output of a command is zero, the negative flag is set when the highest bit is set.

Output

Each instruction can emit an output value. The output can be used in the following ways:

# Output Effect
0 discard The output value is discarded.
1 push The output is pushed to the stack.
2 jump The code pointer is set to the output, thus a jump is taken.
3 jumpr The code pointer is increased by the output, thus a relative jump is taken.

Argument

The instruction argument can provide static input which can be used as a value source for the first input value.

Function Calls

The following chapter defines the SPU Mark I calling convention. It is required that all functions conform to this convention.

To call a function, it is required that the return address is pushed to the stack. After this, a jump is taken to the function address.

call:
	push @returnPoint ; Pushing returnPoint as the return address
	jmp @function     ; Jumps to the function
returnPoint:

SPU Mark I provides the instruction cpget which pushes by default the address of the second next instruction which resembles the code above. This behaviour allows position independent code:

call:
	cpget         ; pushs implicit returnPoint
	jmp @function ; Calls function

Functions can now return by calling ret when the return address is on top of the stack. A simple function that does a system call may look like this:

function:
	syscall
	ret

As most functions utilize local variables, a stack frame is required. Creating this stack frame is done by pushing the current base pointer, then setting the base pointer to the current stack pointer.

enter:
	bpget ; Save current base pointer
	spget ; Get current stack pointer
	bpset ; Set new base pointer

Returning a function with this mechanism is by setting the stack pointer to the current base pointer, then popping the previous base pointer from the stack.

leave:
	bpget ; Get current base pointer
	spset ; Restore stack saved at the beginning
	bpset ; Restore previous base pointer
	ret   ; and jumping back.

This mechanism leaves the base pointer of the calling function intact and also provides a new base pointer for the current function.

TODO

  • 64 Bit arithmetic instructions