Virtual Machine

(1)

www.nand2tetris.org

Building a Modern Computer From First Principles

Virtual Machine

Part II: Program Control

(2)

Where we are at:

Assembler Chapter 6

H.L. Language

&

Operating Sys.

abstract interface

Compiler

Chapters 10 - 11

VM Translator

Chapters 7 - 8

Computer Architecture

Chapters 4 - 5

Gate Logic

Chapters 1 - 3 Electrical

Engineering

Physics Virtual

Machine

abstract interface

Software hierarchy

Assembly Language

abstract interface

Hardware hierarchy

Machine Language

abstract interface

Hardware Platform

abstract interface

Chips &

Logic Gates

abstract interface

Human Thought

Abstract design

Chapters 9, 12

(3)

The big picture

. . .

RISC machine

VM language

other digital platforms, each equipped with its VM implementation RISC

machine language

Hack computer

Hack machine language CISC

machine language

CISC machine

. . .

a high-level^{written in} language

Any computer

. . .

VM implementation

over CISC platforms

VM imp.

over RISC platforms

VM imp.

over the Hack platform VM

emulator Some Other

language

Jack language

Some

compiler Some Other compiler

Jack compiler

. . .

Some

language

. . .

Chapters 1-6

Chapters 7-8

Chapters 9-13

A Java-based emulator is included in the course software suite

Implemented in Projects 7-8

(4)

The VM language

Goal: Complete the specification and implementation of the VM model and language

Method: (a) specify the abstraction (model’s constructs and commands) (b) propose how to implement it over the Hack platform.

Arithmetic / Boolean commands add

sub neg eq gt lt and or not

Memory access commands

pop x (pop into x, which is a variable) push y (y being a variable or a constant)

Program flow commands

label (declaration) goto (label)

if‐goto (label)

Function calling commands

function (declaration)

call (a function)

return (from a function) previous

lecture

this lecture

(5)

The compilation challenge

class Main { static int x;

function void main() {

// Inputs and multiplies two numbers var int a, b, c;

let a = Keyboard.readInt(“Enter a number”);

let b = Keyboard.readInt(“Enter a number”);

let c = Keyboard.readInt(“Enter a number”);

let x = solve(a,b,c);

return;

} }

// Solves a quadratic equation (sort of) function int solve(int a, int b, int c) {

var int x;

if (~(a = 0))

x=(‐b+sqrt(b*b–4*a*c))/(2*a);

else

x=‐c/b;

return x;

} }

Source code (high-level language)

Our ultimate goal:

Translate high-level programs into

executable code.

Compiler

0000000000010000 1110111111001000 0000000000010001 1110101010001000 0000000000010000 1111110000010000 0000000000000000 1111010011010000 0000000000010010 1110001100000001 0000000000010000 1111110000010000 0000000000010001 0000000000010000 1110111111001000 0000000000010001 1110101010001000 0000000000010000 1111110000010000 0000000000000000 1111010011010000 0000000000010010 1110001100000001 0000000000010000 1111110000010000 0000000000010001

...

Target code

(6)

The compilation challenge / two-tier setting

if (~(a = 0))

x = (‐b+sqrt(b*b–4*a*c))/(2*a) else

x = ‐c/b Jack source code

push a push 0 eq

if‐goto elseLabel push b

neg push b push b call mult push 4 push a call mult push c call mult call sqrt add

push 2 push a call mult div

pop x

goto contLable elseLabel:

push c neg push b call div pop x contLable:

Compiler

VM (pseudo) code

0000000000010000 1110111111001000 0000000000010001 1110101010001000 0000000000010000 1111110000010000 0000000000000000 1111010011010000 0000000000010010 1110001100000001 0000000000010000 1111110000010000 0000000000010001 0000000000010000 1110111111001000 0000000000010001 1110101010001000 0000000000010000 1111110000010000 0000000000000000 1111010011010000 0000000000010010 1110001100000001 0000000000010000 1111110000010000 0000000000010001 0000000000010010 1110001100000001 ...

VM translator

Machine code

 We’ll develop the compiler later in the course

 We now turn to describe how to complete the implementation of the VM language

 That is -- how to translate each VM command into assembly

commands that perform the desired semantics.

(7)

// Computes x = (‐b + sqrt(b^2 ‐4*a*c)) / 2*a if (~(a = 0))

x = (‐b + sqrt(b * b – 4 * a * c)) / (2 * a) else

x = ‐ c / b

Typical compiler’s source code input:

The compilation challenge

arithmetic expressions function call and

return logic Boolean

expressions program flow logic

(branching)

How to translate such high-level code into machine language?

 In a two-tier compilation model, the overall translation challenge is broken between a front-end compilation stage and a subsequent back-end translation stage

 In our Hack-Jack platform, all the above sub-tasks (handling arithmetic / Boolean

expressions and program flow / function calling commands) are done by the back-end, i.e. by the VM translator.

(previous lecture) (previous lecture)

(this lecture) (this lecture)

(8)

Lecture plan

if‐goto (label)

call (a function)

lecture

(9)

Program flow commands in the VM language

How to translate these three abstractions into assembly?

 Simple: label declarations and goto directives can be effected directly by assembly commands

 More to the point: given any one of these three VM

commands, the VM Translator must emit one or more assembly commands that effects the same semantics on the Hack platform

 How to do it? see project 8.

label c // label declaration

goto c // unconditional jump to the

// VM command following the label c if‐goto c // pops the topmost stack element;

// if it’s not zero, jumps to the

// VM command following the label c In the VM language, the program flow abstraction is delivered using three commands:

VM code example:

function mult 1 push constant 0 pop local 0 label loop

push argument 0 push constant 0 eq

if‐goto end push argument 0 push 1

sub

pop argument 0 push argument 1 push local 0 add

pop local 0 goto loop label end

push local 0 return

(10)

Lecture plan

if‐goto (label)

call (a function)

lecture

(11)

Subroutines

Subroutines = a major programming artifact

 Basic idea: the given language can be extended at will by user-defined commands ( aka subroutines / functions / methods ...)

 Important: the language’s primitive commands and the user-defined commands have the same look-and-feel

 This transparent extensibility is the most important abstraction delivered by high-level programming languages

 The challenge: implement this abstraction, i.e. allow the program control to flow effortlessly between one subroutine to the other

“A well-designed system consists of a collection of black box modules, each executing its effect like magic”

(Steven Pinker, How The Mind Works)

// Compute x = (‐b + sqrt(b^2 ‐4*a*c)) / 2*a if (~(a = 0))

x = (‐b + sqrt(b * b – 4 * a * c)) / (2 * a) else

x = ‐ c / b

(12)

Subroutines in the VM language

The invocation of the VM’s primitive commands and subroutines

follow exactly the same rules:

The caller pushes the necessary

argument(s) and calls the command / function for its effect

The called command / function is

responsible for removing the argument(s) from the stack, and for popping onto

the stack the result of its execution.

function mult 1 push constant 0

pop local 0 // result (local 0) = 0 label loop

push argument 0 push constant 0 eq

if‐goto end // if arg0 == 0, jump to end push argument 0

push 1 sub

pop argument 0 // arg0‐‐

push argument 1 push local 0 add

pop local 0 // result += arg1 goto loop

label end

push local 0 // push result return

Called code, aka “callee” (example) ...

// computes (7 + 2) * 3 ‐ 5 push constant 7

push constant 2 add

push constant 3 call mult

push constant 5 sub

...

Calling code (example)

VM subroutine call-and-return commands

(13)

Function commands in the VM language

Q: Why this particular syntax?

A: Because it simplifies the VM implementation (later).

function g nVars // here starts a function called g, // which has nVars local variables

call g nArgs // invoke function g for its effect;

// nArgs arguments have already been pushed onto the stack return // terminate execution and return control to the caller

(14)

Function call-and-return conventions

function mult 1 push constant 0

pop local 0 // result (local 0) = 0 label loop

... // rest of code omitted label end

push local 0 // push result return

called function aka “callee” (example) function demo 3

...

push constant 7 push constant 2 add

push constant 3 call mult

...

Calling function

Call-and-return programming convention

 The caller must push the necessary argument(s), call the callee, and wait for it to return

 Before the callee terminates (returns), it must push a return value

 At the point of return, the callee’s resources are recycled, the caller’s state is re-instated, execution continues from the command just after the call

 Caller’s net effect: the arguments were replaced by the return value (just like with primitive commands)

Behind the scene

 Recycling and re-instating subroutine resources and states is a major headache

 Some agent (either the VMor the compiler) should manage it behind the scene “like magic”

 In our implementation, the magic is VM / stack-based, and is considered a great CS gem.

Although not obvious in this example, every VM function has a private set of 5 memory segments (local, argument, this, that, pointer)

These resources exist as long as the function is running.

(15)

The function-call-and-return protocol

The caller’s view:

 When I start executing, my argument segment has been initialized with actual argument values passed by the caller

 My local variables segment has been allocated and initialized to zero

 The static segment that I see has been set to the static segment of the VM file to which I belong, and the working stack that I see is empty

 Before exiting, I must push a value onto the stack and then use the command return.

 Before calling a function g, I must push onto the stack as many arguments as needed by g

 Next, I invoke the function using the command call g nArgs

 After greturns:

 The arguments that I pushed before the call have disappeared from the stack, and a return value (that always exists)

appears at the top of the stack

 All my memory segments (local, argument, this, that, pointer) are the same as before the call.

The callee’s (g ‘s) view:

Blue = VM function

writer’s responsibility Black = black box magic,

delivered by the VMimplementation

Thus, the VM implementation writer must worry about the “black operations” only.

function g nVars call g nArgs return

(16)

When function f calls function g, the VM implementation must:

 Save the return address within f ‘s code:

the address of the command just after the call

 Save the virtual segments of f

 Allocate, and initialize to 0, as many local variables as needed by g

 Set the local and argument segment pointers of g

 Transfer control to g.

When g terminates and control should return to f, the VM implementation must:

 Clear g ’s arguments and other junk from the stack

 Restore the virtual segments of f

 Transfer control back to f

(jump to the saved return address).

Q: How should we make all this work “like magic”?

A: We’ll use the stack cleverly.

The function-call-and-return protocol: the VM implementation view

function g nVars call g nArgs return

(17)

The implementation of the VM’s stack on the host Hack RAM

Global stack:

the entire RAM area dedicated for holding the stack

Working stack:

The stack that the current function sees

 At any point of time, only one function (the current function)

is executing; other functions may be waiting up the calling chain

 Shaded areas: irrelevant to the current function

 The current function sees only the working stack, and has access only to its memory segments

 The rest of the stack holds the frozen states of all the functions up the calling hierarchy.

(18)

Implementing the call g nArgs command

Implementation: If the VM is implemented as a program that translates VM code into assembly code, the translator must emit the above logic in assembly.

// In the course of implementing the code of f

// (the caller), we arrive to the command call g nArgs.

// we assume that nArgs arguments have been pushed // onto the stack. What do we do next?

// We generate a symbol, let’s call it returnAddress;

// Next, we effect the following logic:

push returnAddress // saves the return address push LCL // saves the LCL of f

push ARG // saves the ARG of f push THIS // saves the THIS of f push THAT // saves the THAT of f ARG = SP‐nArgs‐5 // repositions SP for g LCL = SP // repositions LCL for g goto g // transfers control to g returnAddress: // the generated symbol

call g nArgs

None of this code is executed yet ...

At this point we are just generating code (or simulating the VM code on some platform)

(19)

Implementing the function g nVars command

argument nArgs-1 ARG

saved THIS saved ARG saved returnAddress

saved LCL

local 0 local 1

. . .

local nVars-1 argument 0 argument 1

. . .

frames of all the functions up the calling chain

LCL

SP

saved THAT

function g nVars

// to implement the command function g nVars, // we effect the following logic:

g:

repeat nVars times:

push 0

(20)

Implementing the return command

// In the course of implementing the code of g, // we arrive to the command return.

// We assume that a return value has been pushed // onto the stack.

// We effect the following logic:

frame = LCL // frame is a temp. variable retAddr = *(frame‐5) // retAddr is a temp. variable

*ARG = pop // repositions the return value // for the caller

SP=ARG+1 // restores the caller’s SP THAT = *(frame‐1) // restores the caller’s THAT THIS = *(frame‐2) // restores the caller’s THIS ARG = *(frame‐3) // restores the caller’s ARG LCL = *(frame‐4) // restores the caller’s LCL goto retAddr // goto returnAddress

return

(21)

Bootstrapping

SP = 256 // initialize the stack pointer to 0x0100 call Sys.init // call the function that calls Main.main A high-level jack program (aka application) is a set of class files.

By a Jack convention, one class must be called Main, and this class must have at least one function, called main.

The contract: when we tell the computer to execute a Jack program, the function Main.main starts running

Implementation:

 After the program is compiled, each class file is translated into a .vm file

 The operating system is also implemented as a set of .vm files (aka “libraries”) that co-exist alongside the program’s .vm files

 One of the OS libraries, called Sys.vm, includes a method called init.

The Sys.init function starts with some OS initialization code (we’ll deal with this later, when we discuss the OS), then it does call Main.main

 Thus, to bootstrap, the VM implementation has to effect (e.g. in assembly), the following operations:

(22)

 Extends the VM implementation described in the last lecture (chapter 7)

 The result: a single assembly program file with lots of agreed-upon symbols:

VM implementation over the Hack platform

(23)

Proposed API

(24)

Perspective

Benefits of the VM approach

 Code transportability: compiling for

different platforms requires replacing only the VM implementation

 Language inter-operability: code of multiple languages can be shared using the same VM

 Common software libraries

 Code mobility: Internet

 Some virtues of the modularity implied by the VM approach to program translation:

 Improvements in the VM

implementation are shared by all compilers above it

 Every new digital device with a VM implementation gains immediate access to an existing software base

 New programming languages can be implemented easily using simple compilers

. . .

VM language

RISC machine

language Hack

CISC machine

language . . . a high-level^{written in}

language

. . .

VM implementation

over CISC platforms

VM imp.

over RISC

platforms _emulator^VM Translator

Some Other

language Jack

Some

compiler Some Other

compiler compiler

. . .

Some

language . . .

Benefits of managed code:

 Security

 Array bounds, index checking, …

 Add-on code

 Etc.

VM Cons

 Performance.