slike bài giảng môn chương trình dịch chương 9code generation

Java Bytecode AssemblerJava Byte code *.class... • Java source code to Java byte code javac –g -g: to generate debug information • Java byte code to Jasmin code java JasminVisitor Read

Code Generation

Dr Nguyen Hua Phung

Faculty of CSE

HCMUT

Java Bytecode Assembler

Java Byte code (*.class)

source code - Java mapping

• A source program ⇒ Java class(es)

• A global variable ⇒ a static field

• A local variable ⇒ a local variable

• An expression ⇒ an expression

(except an expression statement)

• An invocation ⇒ a static invocation

Some issues

• An array declaration ⇒ a declaration + code

– global ⇒ code in the class init

– local ⇒ code in the enclosing method

• A main method ⇒ a main method with different signature

int main() ⇒ void main(String [] args)

return <integer expr> ⇒ return

public class Cout {

static int a,b;

static float c [] = new float[3 ];

static int foo(int a,int b[]) { boolean c;

• Java source code to Java byte code

javac –g <java source code>

-g: to generate debug information

• Java byte code to Jasmin code

java JasminVisitor <java byte code>

Read JavaToJasmin script for details

• Jasmin code to Java byte code

java -jar jasmin.jar <Jasmin code>

Read RunJasmin script for details

Example –Jasmin

public class Cout {

static int a,b;

static float c[] = new float[3];

static int foo(int a,int b[]) {

public static void main(String[] arg) {

float d[] = new float[3];

.field static b I field static c [F method static <clinit>()V limit stack 1

.limit locals 0 line 3

iconst_3 newarray float putstatic Cout.c [F return

.end method

Example –Jasmin (cont’d)

public class Cout {

static int a,b;

static float c[] = new float[3];

static int foo(int a,int b[]) {

public static void main(String[] arg) {

float d[] = new float[3];

Label0:

.line 1 aload_0 invokespecial java/lang/Object/<init>()V Label1:

return

.end method

Example –Jasmin (cont’d)

public class Cout {

static int a,b;

static float c[] = new float[3];

static int foo(int a,int b[]) {

public static void main(String[] arg) {

float d[] = new float[3];

.line 6 aload_1 iconst_1 iaload iload_0 if_icmple Label0 iconst_1

goto Label1 Label0:

iconst_0 Label1:

istore_2

Label7:

.line 7

iload_2 ifeq Label2 iload_0 ireturn Label2:

.line 8

aload_1 iconst_0 iaload Label4:

ireturn limit stack 2

.limit locals 3

Example –Jasmin (cont’d)

public class Cout {

static int a,b;

static float c[] = new float[3];

static int foo(int a,int b[]) {

public static void main(String[] arg) {

float d[] = new float[3];

.var 1 is d [F from Label3 to Label2 Label1:

.line 11 iconst_3 newarray float astore_1

Label3:

.line 12 getstatic Cout.c [F iconst_0

aload_1 iconst_0

Label0:

.line 14 getstatic Cout.a I iconst_1

iadd putstatic Cout.a I line 15

getstatic Cout.a I bipush 10

if_icmple Label0 Label2:

.line 16 return

Code Generation Design

Independent-machine Code Generation

Dependent-machine Code Generation

Intermediate Code Generation

• Depend on both language and machine

• Select instructions

• Select data objects

• Simulate the execution of the machine

– emitICONST → push()

– emitISTORE → pop()

emitREADVAR

emitILOADemitFLOAD

…(a)

(index)(index)getstatic a …

• Tools are used to manage information used to generate code for a

method

– isMain: generating code for main is different than doing for a normal method

– Labels: are valid in the body of a method

• getNewLabel(): return a new label

• getStartLabel(): return the beginning label of a scope

• getEndLabel(): return the end label of a scope

• getContinueLabel(): return the label where a continue should come

• getBreakLabel(): return the label where a break should come– Local variable array

• getNewIndex(): return a new index for a variable

• getMaxIndex(): return the size of the local variable array– Operand stack

• push(): simulating a push execution

• pop(): simulating a pop execution

• getMaxOpStackSize(): return the max size of the operand stack

• enterScope()

• exitScope()

• enterLoop()

• exitLoop()

Independent-Machine Code

Generation

• Based on the source language

• Use facilities of Frame(.java) and

Intermediate Code Generation

(Emitter.java)

Symbol Entries

• Token id;

• Type type;

• int scope; //GLOBAL or LOCAL

- the scope where the id is declared

• Object object;

- the index if id is a local variable

- the class name if id is a method name

Variable Declarations

• Global variables

.field static name type-desc

- for an array, generating code to initialize the array in the class init

• Local variables

.var var-index is name type-desc scopeStart-label scopeEnd-label

- for an array, generating code to initialize the array in the method

• How ?

– Emitter.emitVARDECL(SymEntry sym)

– Put an array declaration to a list and pass the list to

Emitter.emitCLINIT(<list>) , for global declarations /*TODO*/, or Emitter.emitLISTARRAY(<list>), for locals

Global Procedure Declarations

visitCallExprAST

imulidivfmulfdiv{left for you}

else … return t1;

} Tool: Emitter.java

emitADDOP(String,Type) => iadd, isub,fadd,fsub

emitMULOP(String,Type) => imul,idiv,fmul,fdiv

Type t1=(Type) Type t2=(Type)

• 8 * 10 – 12 / 4 bipush 8bipush 10

imulbipush 12iconst_4idiv

Boolean Expressions

Object visitBinExprAST(ast,…) {

… else if (op.kind == Token.ANDOP) em.printout(em.emitANDOP());

else if (op.kind == Token.OROP) em.printout(em.emitOROP());

• (false && true)

• boolean foo() {putString(“true”); return true;}

(false && foo());

⇒ in short-circuit evaluation,

no string is printed out

iconst_0iconst_1iand

iconst_0invokestatic Cout/foo()Z

“true” is printed out

Use tool Emitter.ANDOP() and Emitter.OROP()

Short-circuit Evaluation

• evaluation order is from left to right

• for andop , if left operand is false, the expr is false

• for orop , if left operand is true, the expr is true

• boolean foo() {putString(“true”); return true;}

ifeq LabelFinvokestatic Cout/foo()Zifeq LabelF

iconst_1goto LabelOLabelF:

Write your own Emitter.emitANDOP(

Relational Expressions

a > b

iload_1 ; aiload_2 ; bif_icmple LabelFiconst_1

goto LabelOLabelF:

iconst_0LabelO:

fload_1 ; afload_2 ; bfcmpl

ifle LabelFiconst_1goto LabelOLabelF:

iconst_0LabelO:

Write your own Emitter.emitRELOP(String,Type,LabelF,LabelO)

istore_1; a

bipush 8bipush 12imul

dupistore_1; adup

istore_2; bbipush 8

bipush 12dup

istore_1; aimul ; b

{E.code = E1.code+Emitter.emitDUP()+T.code;}

bipush 8bipush 12imul

dupistore_1;aistore_2;b

{left for you}

• a + 8

• visitVarExprAST(ast,…) {

SymEntry s=sym.lookup(ast.name.Lexeme); em.printout(em.emitREADVAR(s));

Type t = s.getType();

return t;

}

iload_1 ; abipush 8iadd

class VarExprAST{

Token name; }

Tool:

Emitter.emitREADVAR(SymEntry)

• method-name = class-name + ‘/’ + name.Lexeme

• class name of

iload_1iload_2iconst_4imul

invokestatic Cout/foo …

class CallExprAST {ExprListAST e;Token name;

– Repeat statement (left for you)

– For statement (left for you)

• Continue statement

• Break statement

visitAssiStmtASTvisitCompStmtASTvisitIfThenStmtAST,visitIfThenElseStmtAST

visitWhileStmtASTvisitRepeatStmtASTvisitForStmtASTvisitContStmtASTvisitBreakStmtAST

Assignment Statement

• a = 8 * 12;

• visitAssiStmtAST(ast,…){

ast.l.visit(this,true) ast.e.visit(this,o);

ast.l.visit(this,false)

• b[1] = 8 * 12 ; ???

bipush 8bipush 12imul

istore_1; a

aload_2iconst_1bipush 8bipush 12imul

iastore

class AssiStmtAST {LvalueAST l;

ExprAST e;

abstract class LvalueAST

class VarExprAST extends LvalueAST

Lvalue v.s Rvalue

a = a + 1 visitVarExprAST(ast, o)

o is null → rvalue, use slide 32

o is boolean → lvalue

o is true → do nothing

o is false → use Emitter.emitWRITEVAR

.var 2 is b Z from Label5 to Label6 Label5:

iconst_1 istore_2

index of local variable is reset when parsing out of a scope

Compound Statements (cont’d)

.var 2 is a I from Label2 to Label3

create 2 new labels startLabel and endLabel put these labels onto stacks startStack and endStack keep the new index of local variables

pop on stacks startStack and endStack restore the new index of local variables

Class CompStmtAST {VarDeclPartAST v;

goto Label2Label1:

iconst_0Label2:

ifeq Label3iconst_1istore_3goto Label4Label3:

…

CE2

S

inc/dec

continue Label

continue Label

E1

While Statements while (a > b) do a = a -1 ;

visitWhileStmtAST(ast

ast.e.visit ast.s.visit

Label1:

iload_1 iload_2 if_icmple Label3 iconst_1

goto Label4 Label3:

iconst_0 Label4:

ifeq Label2 iload_1 iconst_1 isub istore_1 goto Label1

Continue and Break

S → continue;

{em.printout(em.emitGOTO(Frame.getContinueLabel()));}

S → break;

{em.printout(em.emitGOTO(Frame.getBreakLabel()));}

Return Statementsreturn 0;

Intermediate Languages (cont’d)

Semantic Rule Production

Intermediate Languages (cont’d)

t4 = b * t3 t5 = t2 + t4

a = t5

t1 = -c t2 = b * t1 t3 = t2 + t2

a = t3

a = b * -c + b * -c

Intermediate Languages (cont’d)

– jump backward ⇒ easy

– jump forward ⇒ difficult ⇒ backpatching

• makelist(i)

• merge(l1,l2)

• backpatch(l,i)

*R indirect register

1 c+contents(R)

c(R) indexed

0 R

R register

1 M

M absolute

ADDED COSTADDRESS

FORMMODE

(R1,R2 contain values of b and c, respectively,

(R0,R1 and R2 contain addresses of a,b and c, respectively)

cost = 6

cost = 6

cost = 2

Register Allocation

• Register allocation

– which variables will reside in registers at a

point in the program

• Register assignment

– which register a variable will reside in

variable i should reside in a register, r0

variable c should reside in a register, r1

c resides in r0

where is c ?

Basic Blocks

-Algorithm 9.1, page 529

Flow Graphs

• A flow graph is a directed graph, whose

nodes are basic blocks and whose edges

are flow-of-control.

-Next-Use and Liveness

i: x =

…j: … = x

no assignment to xthere is a path from i to j

j uses the value of x computed at i

a variable x is live at point p if the value of x could be used along some

path in the flow graph starting at p Otherwise, it is dead

Read page 534 for details

Read page 633 for details

(1) a = 1

(2) b = a + 1

(3) print(b);

(2) uses the value of a computed at (1)

a is live at (2) but dead at (3)

Simple Code Generator

• For a basic block

• Assume computed results can be left in registers

as long as possible, storing them only

– their register is needed for another computation

– just at the end of a basic block

Simple Code-Generation Algorithm

• getreg() to know the location L where the result should

be stored.

• Consult the address descriptor for y to know y’, where the current value of y is

• If y’ is not L, generate MOV y’,L

• Generate OP z’, L, where z’ is the current location of z.

• Update address descriptor(x) = {L}

• If L is a register, update descriptor(L) = {x}

– has no next uses

– are not live on exit from the block

x := y op z

Simple Function Getreg

return the location L that holds the value of x for the statement x = y op z

1 if y

• is in a register (holds the value of no other names)

• is not live

• has not next use

⇒ return the register as L and descriptor(y) ∉ L

2 else return an empty register if there is one

3 else if x has next use,

or op is an operator that requires a register

⇒ find an occupied register R

store the value of R into a memory location (MOV R,M)update descriptor(M)

return R

d lives at the end while the others are dead

all registers (r0,r1) are empty

Code ???

d = v + u

u in r1

v in r0

r0 contains vr1 contains u

MOV a,r1SUB c,r1

u = a - c

t in r0r0 contains t

MOV a, r0SUB b,r0

t = a - b

Address descriptor

Register descriptorCode generated

Statements