Compiler Presentation

Home Page
Title Page
CS432F/CSL 728: JJ II
Compiler Design J I
July 2004
Page 1 of 100
S. Arun-Kumar Go Back
sak@cse.iitd.ernet.in
Department of Computer Science and Engineering Full Screen
I. I. T. Delhi, Hauz Khas, New Delhi 110 016.

Close
July 30, 2004

Quit
•First •Prev •Next •Last •Go Back •Full Screen •Close •Quit
Contents Home Page
• first The Bigpicture last Title Page
• first Lexical Analysis last JJ II
• first Syntax Analysis last

J I
– first Context-free Grammars last
– first Ambiguity last
Page 2 of 100
– first Shift-Reduce Parsing last
– first Parse Trees last Go Back
• first Semantic Analysis last

Full Screen
• first Synthesized Attributes last
• first Inherited Attributes last Close
Contd . . .
Quit
Home Page
Contents Title Page
. . . Contd JJ II
• first Abstract Syntax Trees last

J I
• first Symbol Tables last
• first Intermediate Representation last Page 3 of 100
– IR: Properties
Go Back
– Typical Instruction Set
– IR: Generation
Full Screen
• first Runtime Structure last

Close
Quit
Home Page
Title Page
JJ II
J I
Page 4 of 100
Go Back
Full Screen
Close
Quit
The Big Picture: 1 Home Page
stream of Title Page

characters
SCANNER
stream of JJ II
tokens
J I
Page 5 of 100
Go Back
Full Screen
Close
Quit

characters
SCANNER
stream of JJ II
tokens
PARSER
parse tree J I
Page 6 of 100
Go Back
Full Screen
Close
Quit

characters
SCANNER
stream of JJ II
tokens
PARSER
parse tree J I
SEMANTIC ANALYZER
abstract
syntax tree Page 7 of 100
Go Back
Full Screen
Close
Quit

characters
SCANNER
stream of JJ II
tokens
PARSER
parse tree J I
SEMANTIC ANALYZER
abstract
I.R. CODE GENERATOR

intermediate Go Back
representation
Full Screen
Close
Quit

characters
SCANNER
stream of JJ II
tokens
PARSER
parse tree J I
SEMANTIC ANALYZER
abstract
I.R. CODE GENERATOR

representation
OPTIMIZER
optimized Full Screen

intermediate
representation
Close
Quit

characters
SCANNER
stream of JJ II
tokens
PARSER
parse tree J I
SEMANTIC ANALYZER
abstract
I.R. CODE GENERATOR

representation
OPTIMIZER

intermediate
representation
CODE GENERATOR Close
target code
Quit

characters
SCANNER
stream of JJ II
tokens
PARSER
parse tree J I
ERROR− SEMANTIC ANALYZER SYMBOL
abstract
HANDLER TABLE Page 11 of 100
syntax tree
MANAGER
I.R. CODE GENERATOR
representation
OPTIMIZER

intermediate
representation
CODE GENERATOR Close
target code
Quit
The Big Picture: 7
Home Page
Title Page
stream of
characters
SCANNER JJ II
stream of
tokens
PARSER
parse tree J I
ERROR− SEMANTIC ANALYZER SYMBOL
abstract
HANDLER TABLE
syntax tree
MANAGER
I.R. CODE GENERATOR Page 12 of 100
intermediate
representation
OPTIMIZER
optimized Go Back
intermediate
representation
CODE GENERATOR
target code Full Screen
Scanner Parser Semantic Analysis Symbol Table

Close
IR Optimization Contents
Quit
Home Page
Title Page
JJ II
J I
Page 13 of 100
Go Back
Full Screen
Close
Quit
Home Page
Scanning: 1 Title Page
JJ II
• Takes a stream of characters and identifies tokens from

the lexemes. J I
Page 14 of 100
• Eliminates comments and redundant whitepace.

Go Back
• Keeps track of line numbers and column numbers and Full Screen
passes them as parameters to the other phases to en-
able error-reporting to the user.
Close
Quit
Home Page
• Whitespace: A sequence of space, tab, newline, JJ II

carriage-return, form-feed characters etc.
• Lexeme: A sequence of non-whitespace characters de- J I
limited by whitespace or special characters (e.g. oper-

ators like +, -, *). Page 15 of 100
• Examples of lexemes.
Go Back
– reserved words, keywords, identifiers etc.

– Each comment is usually a single lexeme Full Screen
– preprocessor directives
Close
Quit
Home Page
• Token: A sequence of characters to be treated as a

single unit. JJ II
• Examples of tokens.
J I
– Reserved words (e.g. begin, end, struct, if etc.)
– Keywords (integer, true etc.) Page 16 of 100
– Operators (+, &&, ++ etc)

– Identifiers (variable names, procedure names, pa- Go Back
rameter names)
– Literal constants (numeric, string, character con- Full Screen
stants etc.)
– Punctuation marks (:, , etc.) Close
Quit
Home Page
Scanning: 4
Title Page
JJ II
• Identification of tokens is usually done by a Determinis-
tic Finite-state automaton (DFA). J I
• The set of tokens of a language is represented by a
large regular expression. Page 17 of 100
• This regular expression is fed to a lexical-analyser gen-

erator such as Lex, Flex or ML-Lex. Go Back
• A giant DFA is created by the Lexical analyser genera-

Full Screen
tor.
Close
Quit
Scanning: 5
Home Page
Title Page
a−e,g−z
identifier if identifier
f JJ II
0−9,a−z 0−9,a−z
2 3 4 0−9
J I
i z
j−
0−9
h,
5 6
a−
real Page 18 of 100

0−9 0−9
1
0−9
white Go Back
chars
space 7 8 real
( int
14
other
any char Full Screen
* 10 * 11 )
13 9 12 Close
error error comment
Quit
The Big Picture
Syntax Analysis Home Page
Consider the following two languages over an alphabet

A = {a, b}. Title Page
R = {anbn|n < 100} JJ II

P = {anbn|n > 0}
• R may be finitely represented by a regular expression J I
(even though the actual expression is very long).

Page 19 of 100
• However, P cannot actually be represented by a regular
expression
Go Back
• A regular expression is not powerful enough to repre-

sent languages which require parenthesis matching to Full Screen
arbitrary depths.
• All high level programming languages require an under- Close
lying language of expressions which require parenthe-

ses to be nested and matched to arbitrary depth. Quit
CF-Grammars: Home Page
Definition Title Page
A context-free grammar (CFG) G = hN, T , P, Si consists JJ II

of
• a set N of nonterminal symbols, J I
• a set T of terminal symbols or the alphabet,

Page 20 of 100
• a set P of productions or rewrite rules,
• each production is of the form X −→ α, where Go Back
– X ∈ N is a nonterminal and
– α ∈ (N ∪ T )∗ is a string of terminals and nontermi-
Full Screen
nals
Close
• a start symbol S ∈ N .
Quit
CFG: Example Home Page
G = h{S}, {a, b}, P, Si, where S −→ ab and S −→ aSb Title Page
are the only productions in P .

Derivations look like this: JJ II
J I
S ⇒ ab
Page 21 of 100
S ⇒ aSb ⇒ aabb
S ⇒ aSb ⇒ aaSbb ⇒ aaabbb Go Back
L(G), the language generated by G is {anbn|n > 0}.

Full Screen
Actually can be proved by induction on the length and struc- Close
ture of derivations.
Quit
CFG: Empty word Home Page
G = h{S}, {a, b}, P, Si, where S −→ SS | aSb | ε

Title Page
generates all sequences of matching nested parentheses,
including the empty word ε.
JJ II
A leftmost derivation might look like this:

J I
S ⇒ SS ⇒ SSS ⇒ SS ⇒ aSbS ⇒ abS ⇒ abaSb . . .

Page 22 of 100
A rightmost derivation might look like this:

Go Back
S ⇒ SS ⇒ SSS ⇒ SS ⇒ SaSb ⇒ Sab ⇒ aSbab . . .

Other derivations might look like God alone knows what! Full Screen
S ⇒ SS ⇒ SSS ⇒ SS ⇒ . . . Close
Could be quite confusing!

Quit
Home Page
CFG: Derivation trees Title Page
1 JJ II
Derivation sequences
• put an artificial order in which productions are fired. J I
• instead look at trees of derivations in which we may think

of productions as being fired in parallel. Page 23 of 100
• There is then no highlighting in red to determine which

Go Back
copy of a nonterminal was used to get the next member
of the sequence.
Full Screen
• Of course, generation of the empty word ε must be

shown explicitly in the tree. Close
Quit
CFG: Derivation trees
2 Home Page
Title Page
S
JJ II
S S
J I
S S εε Page 24 of 100
Go Back
a S b a S b
Full Screen
Derivation tree of
ε a S b
abaabb Close
ε
Quit
3 Home Page
Title Page
S JJ II
S S
J I
S S a S b Page 25 of 100
Go Back
a S b ε a S b
Full Screen
ε ε
Another
Close
Derivation tree of
abaabb
Quit
4 Home Page
Title Page
S
JJ II
S S
J I
S S a S b Page 26 of 100
ε a b
Go Back
a S b S
Full Screen
ε ε
Yet another
Close
Derivation tree of
abaabb
Quit
Home Page
Title Page
JJ II
J I
Page 27 of 100
Go Back
Full Screen
Close
Quit
Ambiguity: 1
Home Page
E → I | C | E+E | E∗E
I → y | z Title Page
C → 4
Consider the sentence y + 4 ∗ z. JJ II
J I
E E
Page 28 of 100
Go Back
Full Screen
Close
Quit
Ambiguity: 2
Home Page
E → I | C | E+E | E∗E
C → 4
J I
E E
Page 29 of 100
E + E E * E
Go Back
Full Screen
Close
Quit
Ambiguity: 3
Home Page
E → I | C | E+E | E∗E
C → 4
J I
E E
Page 30 of 100
E + E E * E
Go Back
* E E + E
I E I
Full Screen
Close
Quit
Ambiguity: 4
Home Page
E → I | C | E+E | E∗E
C → 4
J I
E E
Page 31 of 100
E + E E * E
Go Back
* E E + E
I E I
Full Screen
I I C
y C z
Close
Quit
Ambiguity: 5
Home Page
E → I | C | E+E | E∗E
C → 4
J I
E E
Page 32 of 100
E + E E * E
Go Back
* E E + E
I E I
Full Screen
I I C
y C z
Close
z y 4
Quit
4
Home Page
Title Page
JJ II
J I
Page 33 of 100
Go Back
Full Screen
Close
Quit
Parsing: 0
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
a − a / b
Page 34 of 100
Go Back
Full Screen
Close
Quit
Parsing: 1
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
− a / b
Page 35 of 100
Principle: Go Back
Reduce
whenever possible.
Full Screen
Shift only when
reduce is
impossible Close
a
Shift Quit
Parsing: 2
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
− a / b
Page 36 of 100
Go Back
Full Screen
Close
D Reduce by r5 Quit
Parsing: 3
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
− a / b
Page 37 of 100
Go Back
Full Screen
Close
T Reduce by r4 Quit
Parsing: 4
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
− a / b
Page 38 of 100
Go Back
Full Screen
Close
E Reduce by r2 Quit
Parsing: 5
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
a / b
Page 39 of 100
Go Back
Full Screen
Close
− Shift
E
Quit
Parsing: 6
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
/ b
Page 40 of 100
Go Back
Full Screen
a
Shift Close
−
E
Quit
Parsing: 7
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
/ b
Page 41 of 100
Go Back
Full Screen
D Reduce by r5 Close
−
E
Quit
Parsing: 8
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
/ b
Page 42 of 100
Go Back
Full Screen
T Reduce by r4 Close
−
E
Quit
Parsing: 8a
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
/ b
Page 43 of 100
Go Back
Full Screen
−
E
Quit
Parsing: 9a
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
/ b
Page 44 of 100
Go Back
Full Screen
Close
E Reduce by r1 Quit
Parsing: 10a
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
b
Page 45 of 100
Go Back
Full Screen
Close
/
Shift
E
Quit
Parsing: 11a
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 46 of 100
Go Back
Full Screen
b Shift Close
/
E
Quit
Parsing: 12a
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 47 of 100
Go Back
Full Screen
/
E
Quit
Parsing: 13a
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 48 of 100
Go Back
Full Screen
T
Reduce by r4 Close
/
E
Quit
Parsing: 14a
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 49 of 100
Go Back
k!
Full Screen
uc
St
E Reduce by r2 Close
/
Get back!
E
Quit
Parsing: 14b
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 50 of 100
Go Back
Full Screen
E Reduce by r2 Close
/
Get back!
E
Quit
Parsing: 13b
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 51 of 100
Go Back
Full Screen
T
Reduce by r4 Close
Get back! /
E
Quit
Parsing: 12b
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 52 of 100
Go Back
Full Screen
Get back! /
E
Quit
Parsing: 11b
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 53 of 100
Go Back
Full Screen
Get back! b Shift Close

/
E
Quit
Parsing: 10b
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
b
Page 54 of 100
Go Back
Full Screen
Close
/
Get back! Shift
E
Quit
Parsing: 9b
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
/ b
Page 55 of 100
Go Back
Full Screen
Get back to
where you Close
once belonged!
E Reduce by r1 Quit
Parsing: 8b
fied
Home Page
o d i
m Principle:
r1. E E − T
Title Page
r2 E T Reduce whenever possible, but
r3 T T / D but depending upon
JJ II
r4 T D
lookahead
r5 D a | b | ( E )
J I
/ b
Page 56 of 100
Shift instead
of reduce here!
Go Back
ce
i f t − redu
Sh ict Full Screen
confl
−
E
Quit
Parsing: 8
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
/ b
Page 57 of 100
Go Back
Full Screen
−
E
Quit
Parsing: 9
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
b
Page 58 of 100
Go Back
Full Screen
/ Shift
T Close
−
E
Quit
Parsing: 10
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 59 of 100
Go Back
b
Shift Full Screen
/
T Close
−
E
Quit
Parsing: 11
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 60 of 100
Go Back
D Reduce by r5
Full Screen
/
T
Close
−
E
Quit
Parsing: 12
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 61 of 100
Go Back
Full Screen
−
E
Quit
Parsing: 13
Home Page
r1. E E − T
Title Page
r2 E T
r3 T T / D
JJ II
r4 T D
r5 D a | b | ( E )
J I
Page 62 of 100
Go Back
Full Screen
Close
E Reduce by r1 Quit
Parse Trees: 0 Home Page
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 63 of 100
Go Back
Full Screen
Close
a − a / b
Quit
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 64 of 100
Go Back
Full Screen
Close
a − a / b
Quit
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 65 of 100
Go Back
Full Screen
Close
a − a / b
Quit
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 66 of 100
Go Back
Full Screen
Close
a − a / b
Quit
Parse Trees: 3a Home Page
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 67 of 100
Go Back
Full Screen
Close
a − a / b
Quit
Parse Trees: 3b Home Page
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 68 of 100
Go Back
Full Screen
Close
a − a / b
Quit
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 69 of 100
Go Back
Full Screen
D D
Close
a − a / b
Quit
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 70 of 100
Go Back
T T
Full Screen
D D
Close
a − a / b
Quit
Parse Trees: 5a Home Page
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 71 of 100
Go Back
T T
Full Screen
D D
Close
a − a / b
Quit
Parse Trees: 5b Home Page
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 72 of 100
Go Back
T T
Full Screen
D D
Close
a − a / b
Quit
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 73 of 100
Go Back
T T
Full Screen
D D D
Close
a − a / b
Quit
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
Page 74 of 100
E
T
Go Back
T T
Full Screen
D D D
Close
a − a / b
Quit
Title Page
r1. E E − T
r4 T D
r2 E T
r3 T T / D r5 D a | b | ( E )
JJ II
J I
E
Page 75 of 100
E
T
Go Back
T T
Full Screen
D D D
Close
a − a / b
Quit
Home Page
Parsing: Summary: 1 Title Page
• All high-level languages are designed so that they may JJ II

be parsed in this fashion with only a single token look-
ahead. J I
• Parsers for a language can be automatically con-

structed by parger-generators such as Yacc, Bison, ML- Page 76 of 100
Yacc.
• Shift-reduce conflicts if any, are automatically detected Go Back
and reported by the parser-generator.

Full Screen
• Shift-reduce conflicts may be avoided by suitably
redesigning the context-free grammar.
Close
Quit
Home Page
Parsing: Summary: 2 Title Page
• Very often shift-reduce conflicts may occur because

JJ II
of the prefix problem. In such cases many parser-
generators resolve the conflict in favour of shifting.
J I
• There is also a possiblility of reduce-reduce conflicts.
This usually happens when there is more than one non-
Page 77 of 100
terminal symbol to which the contents of the stack may
reduce.
Go Back
• A minor reworking of the grammar to avoid redundant

non-terminal symbols will get rid of reduce-reduce con- Full Screen
flicts.
The Big Picture Close
Quit
Semantic Analysis: 1 Home Page
• Every Programming langauge can be used to program Title Page

any computable function, assuming of course, it has
– unbounded memory, and JJ II
– unbounded time
J I
• The parser of a programming language provides the
framework within which the target code is to be gener-
ated. Page 78 of 100
• The parser also provides a structuring mechanism that Go Back

divides the task of code generation into bits and pieces
determined by the individual nonterminals and produc-
Full Screen
tion rules.
• However, contex-free grammars are not powerful Close
enough to represent all computable functions. Exam-

ple, the language {an bn cn |n > 0}. Quit
Home Page
Semantic Analysis: 2
Title Page
JJ II
• There are context-sensitive aspects of a program that

cannot be represented/enforced by a context-free gram- J I
mar definition. Examples include
– correspondence between formal and actual param- Page 79 of 100
eters
– type consistency between declaration and use. Go Back
– scope and visibility issues with respect to identifiers

Full Screen
in a program.
Close
Quit
Home Page
Title Page
JJ II
J I
Page 80 of 100
Go Back
Full Screen
Close
Quit
Synthesized Attributes:
0 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T F
/
Page 81 of 100
n
F
Go Back
( )
E
E T Full Screen
−
T
F
Close
F
n
Quit
1 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T F
/
Page 82 of 100
n
F
Go Back
( )
E
E T Full Screen
− Synthesized Attributes
4 3 2 1
T
F
Close
F
n
Quit
4 n
2 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T F
/
Page 83 of 100
n
F
Go Back
( )
E
E T Full Screen
4 3 2 1
T
F
Close
4 F
n
Quit
4 n
3 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T F
/
Page 84 of 100
n
F
Go Back
( )
E
E T Full Screen
4 3 2 1
4 T
F
Close
4 F
n
Quit
4 n
4 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T F
/
Page 85 of 100
n
F
Go Back
( )
E
4 E T Full Screen
4 3 2 1
4 T
F
Close
4 F
n
Quit
4 n
5 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T F
/
Page 86 of 100
n
F
Go Back
( )
E
4 E T Full Screen
4 3 2 1
4 T
F
Close
4 F
n 1
Quit
4 n
6 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T F
/
Page 87 of 100
n
F
Go Back
( )
E
4 E T Full Screen
4 3 2 1
4 T
F 1
Close
4 F
n 1
Quit
4 n
7 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T F
/
Page 88 of 100
n
F
Go Back
( )
E
4 E T Full Screen
1
4 3 2 1
4 T
F 1
Close
4 F
n 1
Quit
4 n
8 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T F
/
Page 89 of 100
n
F
Go Back
( )
3
E
4 E T Full Screen
1
4 3 2 1
4 T
F 1
Close
4 F
n 1
Quit
4 n
9 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T F
/
Page 90 of 100
n
3
F
Go Back
( )
3
E
4 E T Full Screen
1
4 3 2 1
4 T
F 1
Close
4 F
n 1
Quit
4 n
10 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T 3 F
/
Page 91 of 100
n
3
F
Go Back
( )
3
E
4 E T Full Screen
1
4 3 2 1
4 T
F 1
Close
4 F
n 1
Quit
4 n
11 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T 3 F
/
Page 92 of 100
n 2
3
F
Go Back
( )
3
E
4 E T Full Screen
1
4 3 2 1
4 T
F 1
Close
4 F
n 1
Quit
4 n
12 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T
−
J I
T 3 F 2
/
Page 93 of 100
n 2
3
F
Go Back
( )
3
E
4 E T Full Screen
1
4 3 2 1
4 T
F 1
Close
4 F
n 1
Quit
4 n
13 Home Page
Title Page
E T F E
JJ II
/ ( ) n
T 1
−
J I
T 3 F 2
/
Page 94 of 100
n 2
3
F
Go Back
( )
3
E
4 E T Full Screen
1
4 3 2 1
4 T
F 1
Close
4 F
n 1
Quit
4 n
14 Home Page
Title Page
E T F E 1
JJ II
/ ( ) n
T 1
−
J I
T 3 F 2
/
Page 95 of 100
n 2
3
F
Go Back
( )
3
E
4 E T Full Screen
1
4 3 2 1
4 T
F 1
Close
4 F
n 1
Quit
4 n
An Attribute Grammar Home Page
Title Page
E 1
E0 → E1−T B E0.val := sub(E1.val, T.val)
JJ II
T 1
E → T B E.val := T.val
T 3 F 2
/ J I
F
3
n 2
T0 → T1/F B T0.val := div(T1.val, F.val)
Page 96 of 100
( )
3
E
4 E T 1
T → F B T.val := F.val
− Go Back
4 T
F 1
F → (E) B F.val := E.val

4 F Full Screen
n 1
4 n
F → n B F.val := n.val Close
Quit
Inherited Attributes: 0
C -style declarations generating int x, y, z. Home Page
D → TL T → int | float
L → L,I | I I → x | y | z Title Page
D
JJ II
T L
J I
I
,
int L Page 97 of 100
z
I
Go Back
L ,
y Full Screen
I
D L T I
Close
x y z int x
Quit
D
JJ II
T L
J I
I
,
int int L Page 98 of 100
z
I
Go Back
L ,
y Full Screen
I
D L T I
Close
x y z int x
Quit
, int
D
JJ II
int T L
J I
I
,
z
I
Go Back
L ,
y Full Screen
I
D L T I
Close
x y z int x
Quit
, int int
D
JJ II
int T L int
J I
I
,
z
I
Go Back
L ,
y Full Screen
I
D L T I
Close
x y z int x
Quit
, int int
D
JJ II
int T L int
J I
I int
,
int int L int Page 101 of 100
z
I
Go Back
L ,
y Full Screen
I
D L T I
Close
x y z int x
Quit
, int int
Inherited Attributes: 5 Home Page
D
Title Page
int T L int
JJ II
I int
,
int int L int
J I
z int
I int
L int , Page 102 of 100
y
Go Back
I
D L T I
Full Screen
x y z int x
Close
, int int
Quit
D
JJ II
int T L int
J I
I int
,
z int
I int
Go Back
L int ,
y int Full Screen
I int
D L T I
Close
x y z int x
Quit
, int int
D
JJ II
int T L int
J I
I int
,
z int
I int
Go Back
L int ,
y int Full Screen
I int
D L T I
Close
x y z int x int
Quit
, int int
Home Page
An Attribute Grammar Title Page
D
D → TL B L.in := T.type JJ II
int T L int
,
I int T → int B T.type := int.int
int int L int J I
z int
L int ,
I int
T → float B T.type := float.f loat
Page 105 of 100
y int
I int L0 → L1,I B L1 := L0.in
Go Back
x int L → I B I.in := L.in

Full Screen
I → id B id.type := I.in
Close
Quit
Home Page
Title Page
JJ II
J I
Page 106 of 100
Go Back
Full Screen
Close
Quit
Abstract Syntax: 0
Home Page
E → E−T | T
T → T /F | F Title Page
F → n | (E)
Suppose we want to evaluate an expression (4 − 1)/2. JJ II
What we actually want is a tree that looks like this:

J I
Page 107 of 100
/
Go Back
−− 2 Full Screen
Close
4 1
Quit
Evaluation: 0 Home Page
Title Page
JJ II
/
J I
Page 108 of 100
−− 2
Go Back
Full Screen
4 1
Close
Quit
Title Page
JJ II
/
J I
Page 109 of 100
−− 2
Go Back
Full Screen
4 1
Close
Quit
Title Page
JJ II
/
J I
Page 110 of 100
3 2
Go Back
Full Screen
Close
Quit
Title Page
JJ II
/
J I
Page 111 of 100
3 2
Go Back
Full Screen
Close
Quit
Evaluation: 4
Home Page
Title Page
1 JJ II
J I
Page 112 of 100
Go Back
Full Screen
Close
But what we actually get during parsing is a tree that looks Quit
like . . .
Abstract Syntax: 1
. . . THIS! E
Home Page
Title Page
T
T F JJ II
/
J I
n
F
Page 113 of 100

( )
E
/
Go Back
E T
−
−− n Full Screen
T
F
n n Close
F
n
Quit
Abstract Syntax: 2
Home Page
Title Page
We use attribute grammar rules to construct the abstract
syntax tree (AST)!. JJ II
But in order to do that we first require two procedures for
tree construction.
J I
makeLeaf(literal) : Creates a node with label literal and re-
turns a pointer to it. Page 114 of 100
makeBinaryNode(opr, opd1, opd2) : Creates a node with la-

bel opr (with fields which point to opd1 and opd2) and Go Back
returns a pointer to the newly created node.

Full Screen
Now we may associate a synthesized attribute called ptr
with each terminal and nonterminal symbol which points to
Close
the root of the subtree created for it.
Quit
Abstract Syntax: 3 Home Page
Title Page
E0 → E1−T B E0.ptr := makeBinaryN ode(−, E1.ptr, T.ptr)

JJ II
E → T B E.ptr := T.ptr
J I
T0 → T1/F B T0.ptr := makeBinaryN ode(/, T1.ptr, F.ptr)
Page 115 of 100
T → F B T.ptr := F.ptr
Go Back
F → (E) B F.ptr := E.ptr

Full Screen
F → n B F.ptr := makeLeaf (n.val)

Close
The Big Picture

Quit
Home Page
Symbol Table:1
Title Page
JJ II
• The store house of context-sensitive and run-time infor-
mation about every identifier in the source program.
J I
• All accesses relating to an identifier require to first find
the attributes of the identifier from the symbol table Page 116 of 100
• Usually organized as a hash table – provides fast ac-

cess. Go Back
• Compiler-generated temporaries may also be stored in

Full Screen
the symbol table
Close
Quit
Home Page
Symbol Table:2 Title Page
Attributes stored in a symbol table for each identifier:

JJ II
• type
• size J I
• scope/visibility information
Page 117 of 100
• base address
• addresses to location of auxiliary symbol tables (in case Go Back
of records, procedures, classes)
• address of the location containing the string which ac- Full Screen
tually names the identifier and its length in the string

pool Close
Quit
Home Page
• A symbol table exists through out the compilation and JJ II

run-time.
• Major operations required of a symbol table: J I
– insertion
Page 118 of 100
– search
– deletions are purely logical (depending on scope Go Back
and visibility) and not physical

• Keywords are often stored in the symbol table before Full Screen
the compilation process begins.

Close
Quit
Home Page
Accesses to the symbol table at every stage of the compi-

lation process, JJ II
Scanning: Insertion of new identifiers.

J I
Parsing: Access to the symbol table to ensure that an
operand exists (declaration before use).
Page 119 of 100
Semantic analysis:
• Determination of types of identifiers from declara- Go Back
tions
• type checking to ensure that operands are used in Full Screen
type-valid contexts.
• Checking scope, visibility violations. Close
Quit
Home Page
Title Page
Symbol Table:5 JJ II
IR generation: . Memory allocation and relativea address
calculation. J I
Optimization: All memory accesses through symbol table

Page 120 of 100
Target code: Translation of relative addresses to absolute
addresses in terms of word length, word boundary etc.
Go Back
The Big picture

a
i.e.relative to a base address that is known only at run-time Full Screen
Close
Quit
Intermediate Home Page
Representation Title Page
Intermediate representations are important for reasons of

JJ II
portability.
• (more or less) independent of specific features of the J I
high-level language.
Example. Java byte-code for any high-level language. Page 121 of 100
• (more or less) independent of specific features of any

particular target architecture (e.g. number of registers, Go Back
memory size)
Full Screen
– number of registers
– memory size Close
– word length
Quit
IR Properties: 1
1. It is fairly low-level containing instructions common to Home Page
all target architectures and assembly languages.

How low can you stoop? . . . Title Page
2. It contains some fairly high-level instructions that are

common to most high-level programming languages. JJ II
How high can you rise?

J I
3. To ensure portability
• an unbounded number of variables and memory lo- Page 122 of 100
cations
• no commitment to Representational Issues Go Back
4. To ensure type-safety
Full Screen
• memory locations are also typed according to the

data they may contain, Close
• no commitment is made regarding word boundaries,

and the structure of individual data items. Quit
Next
Home Page
Title Page
IR: Representation? JJ II
• No commitment to word boundaries or byte boundaries
• No commitment to representation of J I
– int vs. float,

Page 123 of 100
– float vs. double,
– packed vs. unpacked, Go Back
– strings – where and how?.

Back to IR Properties:1 Full Screen
Close
Quit
Home Page
IR: How low can you Title Page
stoop? JJ II
• most arithmetic and logical operations, load and store

instructions etc. J I
• so as to be interpreted easily,
Page 124 of 100
• the interpreter is fairly small,
• execution speeds are high, Go Back
• to have fixed length instructions (where each operand

position has a specific meaning). Full Screen
Back to IR Properties:1 Close
Quit
Home Page
IR: How high can you Title Page
rise? JJ II
• typed variables,
• temporary variables instead of registers. J I
• array-indexing,
Page 125 of 100
• random access to record fields,
• parameter-passing, Go Back
• pointers and pointer management

Full Screen
• no limits on memory addresses
Back to IR Properties:1 Close
Quit
A typical instruction
Home Page
set: 1 Title Page
JJ II
Three address code: A suite of instructions. Each instruc-
tion has at most 3 operands.
J I
• an opcode representing an operation with at most 2
operands
Page 126 of 100
• two operands on which the binary operation is per-

formed Go Back
• a target operand, which accumulates the result of the

(binary) operation. Full Screen
If an operation requires less than 3 operands then one or

Close
more of the operands is made null.
Quit
Home Page
Title Page
A typical instruction JJ II
set: 2 J I
• Assignments (LOAD-STORE)
• Jumps (conditional and unconditional) Page 127 of 100
• Procedures and parameters

Go Back
• Arrays and array-indexing
• Pointer Referencing and Dereferencing Full Screen
Close
Quit
Home Page
A typical instruction Title Page
set: 2 JJ II
– x := y bop z, where bop is a binary operation J I
– x := uop y, where uop is a unary operation

Page 128 of 100
– x := y, load, store, copy or register transfer
• Jumps (conditional and unconditional) Go Back

Full Screen
• Pointer Referencing and Dereferencing Close
Quit
Home Page
set: 2 JJ II
• Jumps (conditional and unconditional) J I
– goto L – Unconditional jump, Page 129 of 100
– x relop y goto L – Conditional jump, where

relop is a relational operator Go Back

• Arrays and array-indexing Full Screen
Quit
Home Page
set: 2 JJ II
• Jumps (conditional and unconditional) J I

Page 130 of 100
– call p n, where n is the number of parameters

– return y, return value from a procedures call Go Back
– param x, parameter declaration

Full Screen
Quit
A typical instruction Home Page
set: 2 Title Page
• Assignments (LOAD-STORE) JJ II
• Jumps (conditional and unconditional)

J I
• Arrays and array-indexing Page 131 of 100
– x := a[i] – array indexing for r-value

– a[j] := y – array indexing for l-value Go Back
Note: The two opcodes are different depending on Full Screen

whether l-value or r-value is desired. x and y are always
simple variables
Close
• Pointer Referencing and Dereferencing

Quit
A typical instruction Home Page
set: 2 Title Page
JJ II
• Jumps (conditional and unconditional)
• Procedures and parameters J I

Page 132 of 100
• Pointer Referencing and Dereferencing
– x := ˆy – referencing: set x to point to y Go Back
– x := *y – dereferencing: copy contents of location

pointed to by y into x Full Screen
– *x := y – dereferencing: copy r-value of y into the

location pointed to by x Close
Picture
Quit
Pointers Home Page
Title Page
x y x y
*x *y x := ^y @y *y
JJ II
x y x y
J I
*x @z *z @z
x := *y Page 133 of 100
*z *z
z z Go Back
y y
*y *y
Full Screen
x x
*x := y @z
@z z
z Close
*z *y
Quit
IR: Generation Home Page
• Can be generated by recursive traversal of the abstract

syntax tree. Title Page
• Can be generated by syntax-directed translation as fol-

lows: JJ II
For every non-terminal symbol N in the grammar of the

source language there exist two attributes J I
N.place , which denotes the address of a temporary

Page 134 of 100
variable where the result of the execution of the gen-
erated code is stored
Go Back
N.code , which is the actual code segment generated.
• In addition a global counter for the instructions gener- Full Screen
ated is maintained as part of the generation process.

• It is independent of the source language but can ex- Close
press target machine operations without committing to

too much detail. Quit
Home Page
IR: Infrastructure Title Page
Given an abstract syntax tree T, with T also denoting its

root node. JJ II
T.place address of temporary variable where result of exe-

cution of the T is stored. J I
newtemp returns a fresh variable name and also installs it in

the symbol table along with relevant information Page 135 of 100
T.code the actual sequence of instructions generated for

Go Back
the tree T.
newlabel returns a label to mark an instruction in the gener- Full Screen
ated code which may be the target of a jump.

emit emits an instructions (regarded as a string). Close
Quit
Home Page
IR: Infrastructure Title Page
Colour and font coding of IR code generation JJ II
• Green: Nodes of the Abstract Syntax Tree

J I
• Brown: Characters and strings of the Intermediate
Representation
Page 136 of 100
• Red: Variables and data structures of the language in
which the IR code generator is written Go Back
• blue: Names of relevant procedures used in IR code gen-

eration. Full Screen
• Black: All other stuff.

Close
Quit
IR: Example Home Page
Title Page
E → id B JJ II
J I
E.place := id.place;
E.code := emit()
Page 137 of 100
E0 → E1 − E2 B
Go Back
E0.place := newtemp; Full Screen

E0.code := E1.code
|| E2.code
Close
|| emit(E0.place := E1.place − E2.place)
Quit
IR: Example
Home Page
S → id := E B Title Page
JJ II
S.code := E.code
|| emit(id.place:=E.place)
J I
S0 → while E do S1 B
Page 138 of 100
S0.begin := newlabel; Go Back

S0.af ter := newlabel;
S0.code := emit(S0.begin:) Full Screen
|| E.code
|| emit(if E.place= 0 goto S0.af ter)
Close
|| S1.code
|| emit(gotoS0.begin)
|| emit(S0.af ter:) Quit
IR: Example
Home Page
S → id := E B Title Page
JJ II
S.code := E.code
|| emit(id.place:=E.place)
J I
S0 → while E do S1 B
Page 139 of 100
S0.begin := newlabel; Go Back

S0.af ter := newlabel;
S0.code := emit(S0.begin:) Full Screen
|| E.code
|| emit(if E.place= 0 goto S0.af ter)
Close
|| S1.code
|| emit(gotoS0.begin)
|| emit(S0.af ter:) Quit
Home Page
IR: Generation Title Page

While generating the intermediate representation, it is
sometimes necessary to generate jumps into code that has
JJ II
not been generated as yet (hence the address of the label
is unknown). This usually happens while processing
J I
• forward jumps
• short-circuit evaluation of boolean expressions Page 140 of 100
It is usual in such circumstances to either fill up the empty

Go Back
label entries in a second pass over the the code or through
a process of backpatching (which is the maintenance of
lists of jumps to the same instruction number), wherein the Full Screen
blank entries are filled in once the sequence number of the

target instruction becomes known. Close
Quit
A Calling Chain Home Page
Main program
Title Page
Globals
Procedure P2
Locals of P2 JJ II
Procedure P21
Locals of P21 J I
Body of P21
Call P21
Page 141 of 100
Body of P2
Call P21
Go Back
Procedure P1
Locals of P1
Full Screen
Body of P1
Call P2
Close
Main body
Call P1
Quit
Run-time Structure: 1 Home Page
Main program
Title Page
Globals
Procedure P2 JJ II
Locals of P2
J I
Procedure P21
Locals of P21
Page 142 of 100
Body of P21
Body of P2 Go Back
Procedure P1
Full Screen
Locals of P1
Body of P1
Close
Main body
Globals
Quit
Main program
Title Page
Globals
Procedure P2 JJ II
Locals of P2
J I
Procedure P21
Locals of P21
Page 143 of 100
Body of P21
Body of P2 Go Back
Procedure P1 Return address to Main

Full Screen
Locals of P1 Dynamic link to Main
Locals of P1
Body of P1
Static link to Main
Close
Formal par of P1
Main body
Globals
Quit
Main program
Title Page
Globals
Procedure P2 JJ II
Locals of P2
Procedure P21 J I
Locals of P21
Return address to last of P1 Page 144 of 100
Body of P21 Dynamic link to last P1

Locals of P2
Body of P2 Static link to last P1 Go Back
Formal par P2

Full Screen
Locals of P1
Body of P1
Static link to Main
Formal par of P1 Close
Main body
Globals
Quit
Main program
Title Page
Globals
Procedure P2 JJ II
Return address to last of P2
Locals of P2
Dynamic link to last P2
Locals of P21
Procedure P21 J I
Static link last P2
Locals of P21 Formal par P21
Return address to last of P1 Page 145 of 100

Locals of P2
Formal par P2

Full Screen
Locals of P1
Body of P1
Static link to Main
Formal par of P1 Close
Main body
Globals
Quit

Main program Dynamic link to last P21 Title Page
Locals of P21
Globals Static link to last P2
Formal par P21
Procedure P2
JJ II
Locals of P2
Dynamic link to last P2
Locals of P21 J I
Procedure P21 Static link last P2
Locals of P21 Formal par P21
Page 146 of 100
Locals of P2
Formal par P2

Full Screen
Locals of P1
Body of P1
Static link to Main
Close
Formal par of P1
Main body
Globals
Quit
Home Page
Title Page
JJ II
J I
Page 147 of 100
Go Back
Full Screen
Close
Quit

Compiler Presentation

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Compiler Presentation

Uploaded by

Copyright:

Available Formats

Home Page

I. I. T. Delhi, Hauz Khas, New Delhi 110 016.

July 30, 2004

• first The Bigpicture last Title Page

• first Lexical Analysis last JJ II

• first Syntax Analysis last

• first Semantic Analysis last

Contents Title Page

• first Abstract Syntax Trees last

• first Runtime Structure last

stream of Title Page

stream of Title Page

stream of Title Page

stream of Title Page

I.R. CODE GENERATOR

stream of Title Page

I.R. CODE GENERATOR

optimized Full Screen

stream of Title Page

I.R. CODE GENERATOR

optimized Full Screen

CODE GENERATOR Close

stream of Title Page

optimized Full Screen

CODE GENERATOR Close

target code Full Screen

Scanner Parser Semantic Analysis Symbol Table

Scanning: 1 Title Page

• Takes a stream of characters and identifies tokens from

• Eliminates comments and redundant whitepace.

Scanning: 2 Title Page

• Whitespace: A sequence of space, tab, newline, JJ II

limited by whitespace or special characters (e.g. oper-

– reserved words, keywords, identifiers etc.

Scanning: 3 Title Page

• Token: A sequence of characters to be treated as a

– Operators (+, &&, ++ etc)

• This regular expression is fed to a lexical-analyser gen-

• A giant DFA is created by the Lexical analyser genera-

real Page 18 of 100

any char Full Screen

error error comment

Consider the following two languages over an alphabet

R = {anbn|n < 100} JJ II

(even though the actual expression is very long).

• A regular expression is not powerful enough to repre-

lying language of expressions which require parenthe-

Definition Title Page

A context-free grammar (CFG) G = hN, T , P, Si consists JJ II

• a set T of terminal symbols or the alphabet,

G = h{S}, {a, b}, P, Si, where S −→ ab and S −→ aSb Title Page

are the only productions in P .

L(G), the language generated by G is {anbn|n > 0}.

Actually can be proved by induction on the length and struc- Close

G = h{S}, {a, b}, P, Si, where S −→ SS | aSb | ε

A leftmost derivation might look like this:

S ⇒ SS ⇒ SSS ⇒ SS ⇒ aSbS ⇒ abS ⇒ abaSb . . .

A rightmost derivation might look like this:

S ⇒ SS ⇒ SSS ⇒ SS ⇒ SaSb ⇒ Sab ⇒ aSbab . . .

Could be quite confusing!

CFG: Derivation trees Title Page

• instead look at trees of derivations in which we may think

• There is then no highlighting in red to determine which