Interpretation (Chapter 8) 1 Course Overview PART I: overview material 1Introduction 2Language processors (tombstone diagrams, bootstrapping) 3Architecture.

1Interpretation (Chapter 8)

Course Overview

PART I: overview material1 Introduction

2 Language processors (tombstone diagrams, bootstrapping)

3 Architecture of a compiler

PART II: inside a compiler4 Syntax analysis

5 Contextual analysis

6 Runtime organization

7 Code generation

PART III: conclusion8 Interpretation

9 Review


Why Interpretation?

• Compiler: Large overhead before the code can be run• Alternative: Direct interpretation of the code (immediate

execution, no time-consuming compilation)• Applications:

– Interactive systems (SQL, shell, etc)

– Simple programming languages (Basic, etc)

– Scripting languages (Perl, Python, etc)

– Programming languages with special requirements (Scheme, Prolog, Smalltalk, etc)

– Write once, run once


Two Kinds of Interpreters

• Iterative interpretation: Well suited for quite simple languages, and fast (at most 10 times slower than compiled languages)

• Recursive interpretation: Well suited for more complex languages, but slower (up to 100 times slower than compiled languages)


Compilation and Interpretation

• Due to the slow speed of recursive interpretation, complex languages (such as Java) are often compiled to simpler languages (such as JVM) that can be interpreted iteratively

TetrisJVMJava

Tetris

x86

Java-->JVM

x86 JVMPPC

PPC

JVMTetris


Iterative Interpretation of Machine Code

• General pattern for iterative interpreters:while (true) {

fetch( );

analyze( );

execute( );

}

• Simulate machine with:– Memory (use arrays for storing code and data)

– I/O (directly)

– CPU (use variables for registers)


Iterative Interpretation of Machine Code

• Fetch: get the next instruction from the code store array at the position pointed to by the instruction pointer; also increment instruction pointer

• Analyze: separate the instruction into an opcode and its operands

• Execute: use a switch statement with one case per each opcode; update memory and registers as specified by the particular instruction


Hypo: a Hypothetical Abstract Machine

• 4096-word code store and 4096-word data store• PC: program counter (register), initially 0• ACC: general purpose accumulator (register), initially 0• 4-bit opcode and 12-bit operand• Instruction set:

Opcode Instruction Meaning0 STORE d word at address d := ACC1 LOAD d ACC := word at address d2 LOADL d ACC := d3 ADD d ACC := ACC + word at address d4 SUB d ACC := ACC – word at address d5 JUMP d PC := d6 JUMPZ d if ACC = 0 then PC := d7 HALT stop execution


Implementation of Hypo in Java

public class HypoInstruction {

public byte op; // opcode field

public short d; // operand field

public static final byte // possible opcodes

STOREop=0, LOADop=1,

LOADLop=2, ADDop =3,

SUBop =4, JUMPop=5,

JUMPZop=6, HALTop=7;

}



public class HypoState {

public static final short CODESIZE=4096;

public static final short DATASIZE=4096;

public HypoInstruction[ ]

code=new HypoInstruction[CODESIZE];

public short[ ] data=new short[DATASIZE];

public short PC;

public short ACC;

public byte status;

public static final byte

RUNNING=0, HALTED=1, FAILED=2;

}



public class HypoInterpreter extends HypoState {public void load( ) {...} // load program into memorypublic void emulate( ) {

PC=0; ACC=0; status=RUNNING;do { // fetch:

HypoInstruction instr=code[PC++];// analyze:byte op=instr.op; byte d=instr.d;// execute:switch (op) { ... // see details on next page}

} while (status==RUNNING);}

}



// execute

switch (op) {

case STOREop: data[d]=ACC; break;

case LOADop: ACC=data[d]; break;

case LOADLop: ACC=d; break;

case ADDop: ACC+=data[d]; break;

case SUBop: ACC-=data[d]; break;

case JUMPop: PC=d; break;

case JUMPZop: if (ACC==0) PC=d; break;

case HALTop: status=HALTED; break;

default: status=FAILED;

}


Iterative Interpretation of Mini-Basic

• Programming languages can be interpreted iteratively unless they have recursive syntactic structures such asCommand ::= if Expression then Command else Command

• EBNF for Mini-Basic:Program ::= Command*

Command ::= Variable = Expression

| read Variable

| write Variable

| go Label

| if Expression RelationalOp Expression

go Label

| stop


Iterative Interpretation of Mini-Basic

• EBNF for Mini-Basic (continued):Expression ::= PrimaryExpression

| Expression ArithmeticOp PrimaryExpressionPrimaryExpression ::= Numeral

| Variable| ( Expression )

ArithmeticOp ::= + | – | * | /RelationalOp ::= = | \= | < | > | =< | >=Variable ::= a | b | c | … | zLabel ::= Digit Digit*Numeral ::=

• The symbol Numeral denotes floating-point literals• 26 predefined variables: a, b, c, …, z


Mini-Basic Interpreter

• Mini-Basic example code:0 read a

1 b=a/2

2 go 4

3 b=(a/b+b)/2

4 d=b*b–a

5 if d>=0 go 7

6 d=0–d

7 if d>=0.01 go 3

8 write b

9 stop


Mini-Basic Interpreter

• Mini-Basic abstract machine:– Data store: array of 26 floating-point values

– Code store: array of commands

– Possible representations for each command:

• Character string (yields slowest execution)

• Sequence of tokens (good compromise)

• AST (yields slowest response time)


Implementing a Mini-Basic Interpreter in Java

class Token {

byte kind;

String spelling;

}

class ScannedCommand {

Token[ ] tokens;

}

public abstract class Command {

public void execute (MiniBasicState state);

}



public class MiniBasicState {

public static final short CODESIZE=4096;

public static final short DATASIZE=26;

public ScannedCommand[ ]

code=new ScannedCommand[CODESIZE];

public float[ ] data=new float[DATASIZE];

public short PC;

public byte status;



}



public class MiniBasicInterpreterextends MiniBasicState {

public void load( ) {...} // load program into memorypublic static Command parse(ScannedCommand scannedCom)

{...} // return a Command ASTpublic void run( ) {

PC=0; status=RUNNING;do { // fetch:

ScannedCommand scannedCom=code[PC++];// analyze:Command analyzedCom=parse(scannedCom);// execute:analyzedCom.execute((MiniBasicState) this);

} while (status==RUNNING);}

}



public class AssignCommand extends Command {byte V; // left sideExpression E; // right sidepublic void Execute(MiniBasicState state) {

state.data[V]=E.evaluate(state);}

}

public class GoCommand extends Command {short L; // destination labelpublic void Execute(MiniBasicState state) {

state.PC=L;}

}

// ReadCommand, WriteCommand, IfCommand, StopCommand, Expression, etc.


Recursive Interpretation

• Recursively defined languages cannot be interpreted iteratively (fetch-analyze-execute), because each command can contain any number of other commands

• Both analysis and execution must be recursive (similar to the parsing phase when compiling a high-level language)

• Hence, the entire analysis must precede the entire execution:– Step 1: Fetch and analyze (recursively)– Step 2: Execute (recursively)

• Execution is a traversal of the decorated AST, hence we can use a new visitor class

• Values (variables and constants) are handled internally


Recursive Interpretation of Mini-Triangle

public abstract class Value { }

public class IntValue extends Value {

public short i;

}

public class BoolValue extends Value {

public boolean b;

}

public class UndefinedValue extends Value {

}



public class MiniTriangleState {

public static final short DATASIZE=...;

Program program; // code store is the decorated AST

Value[ ] data=new Value[DATASIZE];

public byte status;



}



public class MiniTriangleProcessor extends MiniTriangleState implements Visitor {

public void fetchAnalyze( ) {

Parser parser=new Parser(...);

Checker checker=new Checker(...);

StorageAllocator allocator=

new StorageAllocator( );

program=parser.parse( );

checker.check(program);

allocator.allocateAddresses(program);

}

public void run( )

{ program.C.visit(this,null); }

}



public Object visitIfCommand(

IfCommand com, Object arg) {

BoolValue val=(BoolValue)

com.E.visit(this,null);

if (val.b) com.C1.visit(this,null);

else com.C2.visit(this,null);

return null;

}



public Object visitConstDeclaration(

ConstDeclaration decl,

Object arg) {

KnownAddress entity=(KnownAddress)

decl.entity;

Value val=(Value)decl.E.visit(this,null);

data[entity.address]=val;

return null;

}


Case Study: TAM Interpreter

• Variable for each register• Array for code store (Instruction data type)• Array for data store (short type; used for both stack and heap)• Iterative interpretation similar to Hypo• Addressing:

private static short relative (short d, byte r) {

switch(r) {case SBr: return d+SB;case LBr: return d+LB;case L1r: return d+data[LB];case L2r: return d+data[data[LB]];...

}


Case Study: TAM Interpreter

• Use of addressing:switch(op) {

case LOADop: { // push onto stackshort addr=relative(d,r);data[ST++]=data[addr];break; }

case STOREop: { // pop from stackshort addr=relative(d,r);data[addr]=data[––ST];break; }

...}


Usage of the TAM Interpreter

• First write a Triangle program. Assume it is stored in a file “example.tri” within the same folder that contains the Triangle and TAM subfolders.

• Next compile your Triangle program. The command shown below produces an equivalent TAM program in the default file “obj.tam”:

java Triangle/Compiler example.tri• To run this TAM program:

java TAM/Interpreter obj.tam• To view the TAM program in human-readable form:

java TAM/Disassembler obj.tam


For further information

• More details about all these interpreters (Hypo, Mini-Basic, Mini-Triangle, TAM) can be found in the textbook

Interpretation (Chapter 8) 1 Course Overview PART I: overview material 1Introduction 2Language processors (tombstone diagrams, bootstrapping) 3Architecture.

Documents