The Design and Implementation of Gnu Compiler Collectionuday/courses/cs715-08/gcc-intro-1.pdf ·...

The Design and Implementation of GnuCompiler Collection

Uday Khedker

(www.cse.iitb.ac.in/̃ uday)

Department of Computer Science and Engineering,

Indian Institute of Technology, Bombay

July 2008

Part 1

Outline

cs 715 GCC: Outline 1/22

Outline

• Open Source Vs. Free Software

• An Introduction to Compilation

• An overview of Gnu Compiler Collection

• Our work in GCC

• Conclusions

UPK IIT Bombay

Part 2

Open Source Vs. Free Software

cs 715 GCC: Open Source Vs. Free Software 2/22


• The open source initiative: (http://www.opensource.org/)

Emphasis on development methodology

• The Free Software Foundation: (http://www.fsf.org/)

Emphasis on freedom of the user

• In some cases, open source software has restricted user freedom

UPK IIT Bombay



• The open source initiative: (http://www.opensource.org/)

Emphasis on development methodology

• The Free Software Foundation: (http://www.fsf.org/)

Emphasis on freedom of the user

• In some cases, open source software has restricted user freedom

◮ May cooperate with restrictive systems◮ eg. DRM (Digital Rights Management)

UPK IIT Bombay


Open Source and Free Software Development Model

• The Cathedral and the Bazaar

Eric S Raymond, 1997.

• Cathedral: Total Centralized Control

Design, implement, test, release

• Bazaar: Total Decentralization

Release early, release often, let users fix bugs

UPK IIT Bombay


The Bazaar Approach


• Brooks’ law (The Mythical Man Month, 1975)

UPK IIT Bombay


The Bazaar Approach



◮ 12 man month effort

UPK IIT Bombay


The Bazaar Approach



◮ 12 man month effort◮ 1 person working for 12 months

OR12 persons working for 1 month?

UPK IIT Bombay


The Bazaar Approach





• Bazaar approach believes that the two somewhat equivalent ininternet-based distributed development.

UPK IIT Bombay


The Bazaar Approach






• “Given enough eyeballs, all bugs are shallow”.

Code errors, logical errors, and architectural errors.

UPK IIT Bombay


The Bazaar Approach






• “Given enough eyeballs, all bugs are shallow”.

Code errors, logical errors, and architectural errors.

A combination of the two seems more sensible

UPK IIT Bombay

Part 3

Introduction to Compilation

cs 715 GCC: Introduction to Compilation 5/22

Implementation Mechanisms

Source Program

Translator

Target Program

Machine

UPK IIT Bombay



Source Program

Translator

Target Program

Machine

Input Data

UPK IIT Bombay



Source Program

Translator

Target Program

Machine

Input Data

Source Program

Interpreter

Machine

UPK IIT Bombay


Implementation Mechanisms as “Bridges”

• “Gap” between the “levels” of program specification and execution

Program Specification

Machine

UPK IIT Bombay





Machine

Translation

UPK IIT Bombay





Machine

Translation Interpretation

UPK IIT Bombay





Machine

Translation Interpretation

State : VariablesOperations: Expressions,

Control Flow

State : Memory,Registers

Operations: MachineInstructions

UPK IIT Bombay


High and Low Level Abstractions

• A source statement

a = b < 10 ? b : c;

• Spim assembly equivalent

lw $t0, 4($fp) # $t0 <- b

slti $t0, $t0, 10 # $t0 <- $t0 < b

not $t0, $t0 # $t0 <- ! $t0

bgtz $t0, L0: # if $t0 >= 0 goto L0:

lw $t0, 4($fp) # $t0 <- b

b L1: # goto L1:

L0: lw $t0, 8($fp) # L0: $t0 <- c

L1: sw 0($fp), $t0 # L1: a <- $t0

UPK IIT Bombay


Binding

Conceptualisation Coding Compiling Linking Loading Execution

Time

No.ofunboundobjects

Overall Strategy, algorithm, data structure etc

UPK IIT Bombay


Binding


Time

No.ofunboundobjects

Functions, variables, their types etc

UPK IIT Bombay


Binding


Time

No.ofunboundobjects

Machine instructions registers etc

UPK IIT Bombay


Binding


Time

No.ofunboundobjects

Virtual addresses of global variables, functions etc.

UPK IIT Bombay


Binding


Time

No.ofunboundobjects

Actual addresses of code and data

UPK IIT Bombay


Binding


Time

No.ofunboundobjects

Values of variables

UPK IIT Bombay



• Translation = Analysis + Synthesis

Interpretation = Analysis + Execution

UPK IIT Bombay





• Translation InstructionsEquivalent

Instructions

UPK IIT Bombay





• Translation InstructionsEquivalent

Instructions

Interpretation InstructionsActions Implied

by Instructions

UPK IIT Bombay


Language Implementation Models

Analysis

Synthesis

Execution

Compilation

Interpretation

UPK IIT Bombay


Language Processor Models

C,C++

Java,C#

FrontEnd

Optimizer

BackEnd

VirtualMachine

UPK IIT Bombay


Translation Sequence in Our Example: Parsing

a=b<10?b:c;

Input

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree

Issues:

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree

Issues:

• Grammar rules, terminals, non-terminals

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree

Issues:


• Order of application of grammar rules

eg. is it (a = b<10?) followed by (b:c)?

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree

Issues:


• Order of application of grammar rules

eg. is it (a = b<10?) followed by (b:c)?

• Values of terminal symbols

eg. string “10” vs. integer number 10.

UPK IIT Bombay


Translation Sequence in Our Example: Semantic Analysis

a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree

=name(a,int) ?: (int)

<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)

Abstract Syntax Tree(with attributes)

Issues:

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


Issues:

• Symbol tables

Have variables been declared? What are their types?What is their scope?

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


Issues:

• Symbol tables

Have variables been declared? What are their types?What is their scope?

• Type consistency of operators and operands

The result of computing b<10? is bool and not int

UPK IIT Bombay


Translation Sequence in Our Example: IR Generation

a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

Issues:

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

Issues:

• Convert to maximal trees which can beimplemented without altering control flow

Simplifies instruction selection and scheduling,register allocation etc.

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

Issues:

• Convert to maximal trees which can beimplemented without altering control flow

Simplifies instruction selection and scheduling,register allocation etc.

• Linearise control flow by flattening nestedcontrol constructs

UPK IIT Bombay


Translation Sequence in Our Example: Instruction Selection

a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

T0 ← bT0 ← T0 < bT0 ← ! T0

if T0 ≥ 0 goto L0:T1 ← bgoto L1:

L0: T1 ← cL1: a ← T1

Instruction List Issues:

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

T0 ← bT0 ← T0 < bT0 ← ! T0


L0: T1 ← cL1: a ← T1


• Cover trees with as fewmachine instructions aspossible

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

T0 ← bT0 ← T0 < bT0 ← ! T0


L0: T1 ← cL1: a ← T1


• Cover trees with as fewmachine instructions aspossible

• Use temporaries and localregisters

UPK IIT Bombay


Translation Sequence in Our Example: Emitting Instructions

a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

T0 ← bT0 ← T0 < bT0 ← ! T0


L0: T1 ← cL1: a ← T1

Instruction List

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

T0 ← bT0 ← T0 < bT0 ← ! T0


L0: T1 ← cL1: a ← T1

Instruction List

lw $t0, 4($fp)slti $t0, $t0, 10not $t0, $t0bgtz $t0, L0:lw $t0, 4($fp)b L1:

L0: lw $t0, 8($fp)L1: sw 0($fp), $t0

Assembly Code

Issues:

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

T0 ← bT0 ← T0 < bT0 ← ! T0


L0: T1 ← cL1: a ← T1

Instruction List


L0: lw $t0, 8($fp)L1: sw 0($fp), $t0

Assembly Code

Issues:

• Offsets of variables in thestack frame

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

T0 ← bT0 ← T0 < bT0 ← ! T0


L0: T1 ← cL1: a ← T1

Instruction List


L0: lw $t0, 8($fp)L1: sw 0($fp), $t0

Assembly Code

Issues:


• Actual register numbersand assembly mnemonics

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

T0 ← bT0 ← T0 < bT0 ← ! T0


L0: T1 ← cL1: a ← T1

Instruction List


L0: lw $t0, 8($fp)L1: sw 0($fp), $t0

Assembly Code

Issues:


• Actual register numbersand assembly mnemonics

• Code to construct anddiscard activation records

UPK IIT Bombay



a=b<10?b:c;

Input

AsgnStmnt

Lhs = E ;

E ? E : E

E < E

name

name name

name numParse Tree


<

(bool)name(b,int)

name(c,int)

name(b,int)

num(10,int)


=

T0<

b 10IfGoto

Not L0:

T0

=

T1 bGoto

L1:=

T1 c

L0:

=

T1a

L1:

Tree List

T0 ← bT0 ← T0 < bT0 ← ! T0


L0: T1 ← cL1: a ← T1

Instruction List


L0: lw $t0, 8($fp)L1: sw 0($fp), $t0

Assembly Code

UPK IIT Bombay


Typical Front Ends

Parser

UPK IIT Bombay


Typical Front Ends

ParserSourceProgram

Scanner

Tokens

UPK IIT Bombay


Typical Front Ends

ParserSourceProgram

Scanner

Tokens

SemanticAnalyzer

AST

ParseTree

AST or Linear IR+

Symbol Table

UPK IIT Bombay


Typical Front Ends

ParserSourceProgram

Scanner

Tokens

SemanticAnalyzer

AST

ParseTree

AST or Linear IR+

Symbol Table

ErrorHandler

SymtabHandler

UPK IIT Bombay


Typical Back Ends

M/c Ind.IR

M/c Ind.Optimizer

− Compile timeevaluations

− Eliminatingredundantcomputations

M/cInd.IR

UPK IIT Bombay


Typical Back Ends

M/c Ind.IR

M/c Ind.Optimizer



M/cInd.IR

CodeGenerator

M/cDep.IR

− Instruction Selection− Local Reg Allocation− Choice of Order of

Evaluation

UPK IIT Bombay


Typical Back Ends

M/c Ind.IR

M/c Ind.Optimizer



M/cInd.IR

CodeGenerator

M/cDep.IR


Evaluation

M/c Dep.Optimizer

Assembly Code

UPK IIT Bombay


Typical Back Ends

M/c Ind.IR

M/c Ind.Optimizer



M/cInd.IR

CodeGenerator

M/cDep.IR


Evaluation

Assembly Code

RegisterAllocator

InstructionScheduler

PeepholeOptimizer

UPK IIT Bombay

Part 4

Introduction to GCC

cs 715 GCC: Introduction to GCC 19/22

The GNU Tool Chain

gcc

Source Program

Target Program

UPK IIT Bombay


The GNU Tool Chain

gcc

Source Program

Target Program

cc1

UPK IIT Bombay


The GNU Tool Chain

gcc

Source Program

Target Program

cc1 cpp

UPK IIT Bombay


The GNU Tool Chain

gcc

Source Program

Target Program

cc1 cpp

as

UPK IIT Bombay


The GNU Tool Chain

gcc

Source Program

Target Program

cc1 cpp

as

ld

UPK IIT Bombay


The GNU Tool Chain

gcc

Source Program

Target Program

cc1 cpp

as

ld

glibc/newlib

UPK IIT Bombay


The GCC Framework

LanguageSpecific Code

Language andMachine

IndependentGeneric Code

MachinedependentGenerator

Code

MachineDescrip-

tionsGCC

Input Language Machine Description

Parser Genericizer GimplifierTree SSAOptimizer

RTLGenerator

OptimizerCode

Generator

cc1/gcc

UPK IIT Bombay


The GCC Framework


Language andMachine



Code

MachineDescrip-

tionsGCC



RTLGenerator

OptimizerCode

Generator

cc1/gcc

Source Program

Assembly Program

UPK IIT Bombay


The GCC Framework


Language andMachine



Code

MachineDescrip-

tionsGCC



RTLGenerator

OptimizerCode

Generator

Selected

cc1/gcc

UPK IIT Bombay


The GCC Framework


Language andMachine



Code

MachineDescrip-

tionsGCC



RTLGenerator

OptimizerCode

Generator

Selected Copied

Copied

cc1/gcc

UPK IIT Bombay


The GCC Framework


Language andMachine



Code

MachineDescrip-

tionsGCC



RTLGenerator

OptimizerCode

Generator

Selected Copied

CopiedGenerated

Generated

cc1/gcc

UPK IIT Bombay


The cc1 Phase Sequence as IR Chain

The GCC Phase Sequence

Generic GIMPLE RTL ASM

UPK IIT Bombay





HLL

Scope ExplicationLower controlflow

Lower procedures& data

Target

UPK IIT Bombay





HLL

Scope ExplicationLower controlflow

Lower procedures& data

TargetOptimization Opportunities

Rescoping SSA Scheduling

UPK IIT Bombay

Part 5

Configuration and Building

cs 715 GCC: Configuration and Building 22/22

Configuring GCC

configure

UPK IIT Bombay


Configuring GCC

configure

config.guess

UPK IIT Bombay


Configuring GCC

configure

config.guess

configure.in

UPK IIT Bombay


Configuring GCC

configure

config.guess

configure.in config/*

UPK IIT Bombay


Configuring GCC

configure

config.guess


config.sub

UPK IIT Bombay


Configuring GCC

configure

config.guess


config.sub

config.log config.cache config.status

UPK IIT Bombay


Configuring GCC

configure

config.guess


config.sub


config.h.in makefile.in

UPK IIT Bombay


Configuring GCC

configure

config.guess


config.sub


config.h.in makefile.in

Makefile config.h

UPK IIT Bombay

Part 6

About The Course

The Design and Implementation of Gnu Compiler Collectionuday/courses/cs715-08/gcc-intro-1.pdf ·...

Documents

Transcript of The Design and Implementation of Gnu Compiler Collectionuday/courses/cs715-08/gcc-intro-1.pdf ·...