• 沒有找到結果。

Compiler II: Code Generation

N/A
N/A
Protected

Academic year: 2022

Share "Compiler II: Code Generation"

Copied!
20
0
0

加載中.... (立即查看全文)

全文

(1)

www.nand2tetris.org

Building a Modern Computer From First Principles

Compiler II: Code Generation

(2)

The big picture

(Chapter 11)

Jack Program

Toke-

nizer Parser

Code Gene -ration

Syntax Analyzer Jack Compiler

VM code XML code

(Chapter 10)

1. Syntax analysis: extracting the semantics from the source code

2. Code generation: expressing the semantics using the target language lecturethis

previous lecture

(3)

Syntax analysis (review)

Class Bar {

method Fraction foo(int y) { var int temp; // a variable let temp = (xxx+12)*‐63;

...

...

<varDec>

<keyword> var </keyword>

<keyword> int </keyword>

<identifier> temp </identifier>

<symbol> ; </symbol>

</varDec>

<statements>

<letStatement>

<keyword> let </keyword>

<identifier> temp </identifier>

<symbol> = </symbol>

<expression>

<term>

<symbol> ( </symbol>

<expression>

<term>

<identifier> xxx </identifier>

</term>

<symbol> + </symbol>

<term>

<int.Const.> 12 </int.Const.> 

</term>

</expression> 

...

Syntax analyzer The code generation challenge:

Program = a series of operations that manipulate data

Compiler: converts each “understood”

(parsed) source operation and data item into corresponding operations and data items in the target language

Thus, we have to generate code for

o handling data

o handling operations

Our approach: morph the syntax analyzer (project 10) into a full-blown compiler: instead of generating XML, we’ll make it generate VM code.

(4)

Memory segments (review)

Where

i

is a non-negative integer and

segment

is one of the following:

static:  holds values of global variables, shared by all functions in the same class argument: holds values of the argument variables of the current function

local: holds values of the local variables of the current function

this: holds values of the private (“object”) variables of the current object that: holds array values (silly name, sorry)

constant: holds all the constants in the range 032767 (pseudo memory segment) pointer: used to anchor this and that to various areas in the heap

temp: fixed 8-entry segment that holds temporary variables for general use;

Shared by all VM functions in the program.

VM memory Commands:

pop

segment i

push

segment i

(5)

VM implementation on the Hack platform (review)

Basic idea: the mapping of the stack and the global segments on the RAM is easy (fixed);

the mapping of the function-level segments is dynamic, using pointers

The stack: mapped on RAM[256 ... 2047];

The stack pointer is kept in RAM address SP static: mapped on RAM[16 ... 255];

each segment reference static i appearing in a VM file named f is compiled to the assembly language symbol f.i (recall that the assembler further

maps such symbols to the RAM, from address 16 onward)

local,argument,this,that: these method-level segments are mapped somewhere from address 2048 onward, in an area called “heap”. The base

addresses of these segments are kept in RAM addresses LCL, ARG, THIS, and THAT. Access to

the i-th entry of any of these segments is implemented by accessing RAM[segmentBase + i]

constant: a truly a virtual segment:

access to constant i is implemented by supplying the constant i.

pointer: discussed later.

Statics

3

12

. . .

4 5

14 15 0 1

13 2 THIS THAT SP LCL ARG

TEMP

255

. . .

16 General

purpose

2047

. . .

256

2048

Stack

. . . Heap

Host

RAM

(6)

Code generation example

method int foo() { var int x;

let x = x + 1;

...

<letStatement>

<keyword> let </keyword>

<identifier> x </identifier>

<symbol> = </symbol>

<expression>

<term>

<identifier> x </identifier>

</term>

<symbol> + </symbol>

<term>

<constant> 1 </constant> 

</term>

</expression>

</letStatement>

Syntax analysis

(note that x is the first local variable declared in the method)

push local 0 push constant 1 add

pop local 0 Code generation

(7)

Handling variables

When the compiler encounters a variable, say x, in the source code, it has to know:

What is x’s data type?

Primitive, or ADT (class name) ?

(Need to know in order to properly allocate RAM resources for its representation)

What kind of variable is x?

local, static, field, argument ?

( We need to know in order to properly allocate it to the right memory segment;

this also implies the variable’s life cycle ).

(8)

Handling variables: mapping them on memory segments

(example)

When compiling this class, we have to create the following mappings:

The class variables nAccounts , bankCommission are mapped on static 0,1

The object fields id, owner, balance are mapped on this 0,1,2

The argument variables sum, bankAccount, when are mapped on arg 0,1,2 The local variables i, j, due are mapped on local 0,1,2.

The target language uses 8 memory segments

Each memory segment, e.g. static,

is an indexed sequence of 16-bit values that can be referred to as

static 0, static 1, static 2, etc.

(9)

Handling variables: symbol tables

How the compiler uses symbol tables:

The compiler builds and maintains a linked list of hash tables, each reflecting a single scope nested within the next one in the list

Identifier lookup works from the current symbol table back to the list’s head

(a classical implementation).

(10)

Handling variables: managing their life cycle

Variables life cycle

static variables: single copy must be kept alive throughout the program duration

field variables: different copies must be kept for each object

local variables: created on subroutine entry, killed on exit

argument variables: similar to local variables.

Good news: the VM implementation already handles all these details !

(11)

class Complex {

// Fields (properties):

int re; // Real part

int im; // Imaginary part ...

/** Constructs a new Complex number */

public Complex (int re, int im) { this.re = re;

this.im = im;

} ...

}

class Foo {

public void bla() { Complex a, b, c;

...

a = new Complex(5,17);

b = new Complex(12,192);

...

c = a; // Only the reference is copied ...

}

Java code

Handling objects: construction / memory allocation

How to compile:

foo = new ClassName(…)  ?

The compiler generates code affecting:

foo = Memory.alloc(n)

Where nis the number of words necessary to represent the object in question, and Memory.alloc is an OS method that returns the base address of a free memory block of size n words.

Following compilation:

(12)

Handling objects: accessing fields

class Complex {

// Properties (fields):

int re; // Real part

int im; // Imaginary part ...

/** Constructs a new Complex number */

public Complex(int re, int im) { this.re = re;

this.im = im;

} ...

/** Multiplies this Complex number by the given scalar */

public void mult (int c) { re = re * c;

im = im * c;

} ...

}

Java code

*(this + 1) = *(this + 1) times

(argument 0) 

How to compile:

im = im * c ?

1. look up the two variables in the symbol table

2. Generate the code:

This pseudo-code should be expressed in the target language.

(13)

Assume that b and r were passed to the function as its first two arguments.

How to compile (in Java):

b.radius = r ?

// Get b's base address:

push argument 0

// Point the this segment to b:

pop pointer 0 // Get r's value push argument 1

// Set b's third field to r:

pop this 2

120 80 radius: 50

x:

y:

color: 3

120 80 50 3012

3013 3014

3 3015

412 3012

...

...

High level program view RAM view

0 ...

b following

compilation b

object

b object (Actual RAM locations of program variables are

run-time dependent, and thus the addresses shown here are arbitrary examples.)

0 0 1

Virtual memory segments just before the operation b.radius=17:

3012 17 0

1

...

...

120 80 17 0

1 2 3012 0

1

3 3012

17 0

1

argument pointer this

...

3

(this0

is now alligned with

RAM[3012])

...

Virtual memory segments just after the operation b.radius=17:

argument pointer this

Handling objects: establishing access to the object’s fields

Background: Suppose we have an object named b of type Ball. A Ball has x,y

coordinates, a radius, and a color.

(14)

Handling objects: method calls

General rule: each method call

foo.bar(v1,v2,...)

is translated into:

push foo push v1 push v2 ...

call bar class Complex {

// Properties (fields):

int re; // Real part

int im; // Imaginary part ...

/** Constructs a new Complex object. */

public Complex(int re, int im) { this.re = re;

this.im = im;

} ...

}

class Foo { ...

public void bla() { Complex x;

...

x = new Complex(1,2);

x.mult(5);

...

} }

Java code

push x push 5 call mult 

How to compile:

x.mult(5)  ?

This method call can also be viewed as:

mult(x,5)

Generate the following code:

(15)

class Bla { ...

void foo(int k) { int x, y;

int[] bar; // declare an array ...

// Construct the array:

bar = new int[10];

...

bar[k]=19;

} ...

Main.foo(2); // Call the foo method ...

Java code

How to compile:

bar = new int(n) ? Generate code affecting:

bar = Memory.alloc(n)

Handling arrays: declaration / construction

19 4315

4316 4317

4324

(bar array)

...

4318

...

...

4315

...

0

bar x y

2 k

(local 0) (local 1) (local 2)

(argument 0) 275

276 277

504

RAM state

...

Following compilation:

(16)

class Bla { ...

void foo(int k) { int x, y;

int[] bar; // declare an array ...

// Construct the array:

bar = new int[10];

...

bar[k]=19;

} ...

Main.foo(2); // Call the foo method ...

Java code

How to compile: bar[k] = 19 ?

// bar[k]=19, or *(bar+k)=19 push bar

push k add

// Use a pointer to access x[k]

pop addr // addr points to bar[k]

push 19

pop *addr // Set bar[k] to 19 VM Code (pseudo)

// bar[k]=19, or *(bar+k)=19 push local 2

push argument 0 add

// Use the that segment to access x[k]

pop pointer 1 push constant 19 pop that 0

VM Code (actual)

19 4315

4316 4317

4324

(bar array)

...

4318

...

...

4315

...

0

bar x y

2 k

(local 0) (local 1) (local 2)

(argument 0) 275

276 277

504

RAM state, just after executing bar[k] = 19

...

Following compilation:

Handling arrays: accessing an array entry by its index

(17)

syntax analysis

parse tree

Handling expressions

((5+z)/-8)*(4^2)

High-level code

push 5 push z add push 8 neg

call div push 4 push 2 call power call mult code

generation

To generate VM code from a parse tree exp, use the following logic:

The codeWrite(exp) algorithm:

if exp is a constant n then output "push n"

if exp is a variable v then output "push v"

if exp is op(exp1) then codeWrite(exp1); output "op";

if exp is (exp1 op exp2) then codeWrite(exp1); codeWrite(exp2); output "op";

if exp is f (exp1, ..., expn) then codeWrite(exp1); ... codeWrite(exp1); output "call f";

VM code

(18)

Handling program flow

if (cond) s1 else

s2 ...

High-level code

VM code to compute and push !(cond) if‐goto L1

VM code for executing s1 goto L2

label L1

VM code for executing s2 label L2

...

VM code

code

generation

while (cond) s

...

High-level code

label L1   

VM code to compute and push !(cond) if‐goto L2

VM code for executing s goto L1

label L2 ...

VM code

code

generation

(19)

Final example

(20)

Perspective

Jack simplifications that are challenging to extend:

Limited primitive type system

No inheritance

No public class fields, e.g. must use r = c.getRadius()

rather than r = c.radius

Jack simplifications that are easy to extend: :

Limited control structures, e.g. no for, switch, …

Cumbersome handling of char types, e.g. cannot use let x=‘c’

Optimization

For example, c=c+1 is translated inefficiently into push c, push 1, add, pop c.

Parallel processing

Many other examples of possible improvements …

參考文獻

相關文件

You are given the wavelength and total energy of a light pulse and asked to find the number of photons it

利用 determinant 我 們可以判斷一個 square matrix 是否為 invertible, 也可幫助我們找到一個 invertible matrix 的 inverse, 甚至將聯立方成組的解寫下.

Then, we tested the influence of θ for the rate of convergence of Algorithm 4.1, by using this algorithm with α = 15 and four different θ to solve a test ex- ample generated as

Numerical results are reported for some convex second-order cone programs (SOCPs) by solving the unconstrained minimization reformulation of the KKT optimality conditions,

Particularly, combining the numerical results of the two papers, we may obtain such a conclusion that the merit function method based on ϕ p has a better a global convergence and

Then, it is easy to see that there are 9 problems for which the iterative numbers of the algorithm using ψ α,θ,p in the case of θ = 1 and p = 3 are less than the one of the

By exploiting the Cartesian P -properties for a nonlinear transformation, we show that the class of regularized merit functions provides a global error bound for the solution of

volume suppressed mass: (TeV) 2 /M P ∼ 10 −4 eV → mm range can be experimentally tested for any number of extra dimensions - Light U(1) gauge bosons: no derivative couplings. =&gt;