1. Compiler 薛智文 TH 6 7 8, DTH 102 cwhsueh@csie.ntu.edu.tw
96 Spring

2. 20:25
/43

3. Why Study Compilers?
Excellent software-engineering example --- theory meets practice.
Essential software tool.
Influences hardware design, e.g., RISC, VLIW.
Tools (mostly “optimization”) for enhancing software reliability and security.
20:25
/43

4. Compilers & Architecture
Modern architectures have very complex structures, especially opportunities for parallel execution.
Sequential programs can only make effective use of these features via an optimizing compiler.
Hardware question: If we implemented this, could a compiler use it?
20:25
/43

5. Software Reliability
Optimization technology (data-flow analysis) used in:
Lock/unlock errors.
Buffers not range-checked.
Memory Leaks.
SQL injection bugs..
20:25
/43

6. What this Course Offers?
Compiler methodology for both compiler implementation and related applications.
Theoretical framework.
Key algorithms.
Hands-on experience.
Nongoal: build a complete optimizing compiler.
20:25
/43

7. Course Outline Part 1 --- Introduction. Part 2 --- Scanner.
Part Parser.
Part Syntax-Directed Translation.
Part Symbol Table.
Part Intermediate Code Generation.
Part Run Time Storage Organization.
Part Optimization.
Part How to Write a Compiler.
Part A Simple Code Generation (PSEUDO) Example.
20:25
/43

8. Introduction Compiler is one of language processors. input
source program
target program
output
20:25
/43

9. What is a Compiler? Definitions: Compiler writing spans: History:
The software system translates description of computations into a program executable by a computer.
Source and target must be equivalent!
Compiler writing spans:
programming languages;
machine architecture;
language theory;
algorithms and data structures;
software engineering.
History:
1950: the first FORTRAN compiler took 18 man-years;
now: using software tools, can be done in a few months as a student’s project.
input
Compiler
source program
target program
output
20:25
/43

10. An Interpreter
input
Interpreter
source program
output
20:25
/43

11. A Hybrid Compiler source program intermediate program Virtual Machine
Translator
intermediate program
Virtual
Machine
output
input
20:25
/43

12. A Language-Processing System
source program
Preprocessor
modified source program
Compiler
target assembly program
Assembler
relocatable machine code
Linker/Loader
target machine code
library files
relocatable object files
20:25
/43

13. Applications Computer language compilers.
Translator: from one format to another.
query interpreter
text formatter
silicon compiler
infix notation  postfix notation:
pretty printers
· · ·
Software productivity tools.
3 + 5 – 6 * 6
* –
20:25
/43

14. Relations with Computational Theory
a set of grammar rules ≡ the definition of a particular machine.
also equivalent to a set of languages recognized by this machine.
a type of machines: a family of machines with a given set of operations, or capabilities;
power of a type of machines
≡ the set of languages that can be recognized by this type of machines.
20:25
/43

15. A Language-Processing System
source program
Preprocessor
modified source program
Compiler
target assembly program
Assembler
relocatable machine code
Linker/Loader
target machine code
library files
relocatable object files
20:25
/43

16. Phases of a Compiler character stream Lexical Analyzer (scanner)
token stream
Machine-Independent Code Optimizer
optimized intermediate representation
Code Generator
relocatable machine code
Machine-Dependent Code Optimizer
target-machine code
Semantic Analyzer
annotated abstract-syntax tree
Syntax Analyzer (parser)
abstract-syntax tree
Intermediate Code Generator
intermediate representation
Error
Handling
Symbol
Table
20:25
/43

17. Lexical Analyzer (Scanner)
Actions:
Reads characters from the source program;
Groups characters into lexemes , i.e., sequences of characters that “go together”, following a given pattern ;
Each lexeme corresponds to a token .
the scanner returns the next token, plus maybe some additional information, to the parser;
The scanner may also discover lexical errors, i.e., erroneous characters.
The definitions of what a lexeme, token or bad character is depend on the definition of the source language.
20:25
/43

18. Scanner Example for C Lexeme: C statement
Symbol Table
Lexeme: C statement
position = initial + rate ＊ 60;
(Lexeme) position = initial rate ＊ ;
< id,1> <=> <id,2> <+> <id,3> <＊> <60> <;>
(Token) ID ASSIGN ID PLUS ID TIME INT SEMI-COL
Arbitrary number of blanks (white spaces) between lexemes.
Erroneous sequence of characters, that are not parts of comments, for the C language:
control characters @
2abc 1
position … 2
initial 3
rate
20:25
/43

19. Syntax Analyzer (Parser)
Actions:
Group tokens into grammatical phrases , to discover the underlying structure of the source
Find syntax errors , e.g., the following C source line:
(Lexeme) index = * ;
(Token) ID ASSIGN INT TIMES SEMI-COL
Every token is legal, but the sequence is erroneous!
May find some static semantic errors , e.g., use of undeclared variables or multiple declared variables.
May generate code, or build some intermediate representation of the source program, such as an abstract-syntax tree.
20:25
/43

20. Parser Example for C Source code: position = initial + rate ＊ 60
< id,1> <=> <id,2> <+> <id,3> <＊> <60>
Abstract-syntax tree:
interior nodes of the tree are OPERATORS;
a node’s children are its OPERANDS;
each subtree forms a logical unit .
the subtree with ＊ at its root shows that ＊ has higher precedence than +, the operation “rate ＊ 60” must be performed as a unit, not “initial + rate”.
Where is ”;”? = + ＊
<id,3>
< id,1>
<id,2> 60
Symbol Table 1
position … 2
initial 3
rate
20:25
/43

21. Semantic Analyzer Actions:
Check for more static semantic errors, e.g., type errors .
May annotate and/or change the abstract syntax tree.
Symbol Table 1
position … 2
initial 3
rate = + ＊
<id,3>
< id,1>
<id,2> 60 = + ＊
<id,3>
< id,1>
<id,2>
int_to_float 60
20:25
/43

22. Intermediate Code Generator
Actions: translate from abstract-syntax trees to intermediate codes.
One choice for intermediate code is 3-address code :
Each statement contains
at most 3 operands;
in addition to “=”, i.e., assignment, at most one operator.
An ”easy” and “universal” format that can be translated into most assembly languages. = + ＊
<id,3>
< id,1>
<id,2>
int_to_float 60
t1 = int_to_float(60)
t2 = id3 ＊ t1
t3 = id2 + t2
id1 = t3
20:25
/43

23. Optimizer Improve the efficiency of intermediate code.
Goal may be to make code run faster , and/or to use the least number of registers · · ·
Current trends:
to obtain smaller, but maybe slower, equivalent code for embedded systems;
to reduce power consumption.
t1 = int_to_float(60)
t2 = id3 ＊ t1
t3 = id2 + t2
id1 = t3
t1 = id3 ＊ 60.0
id1 = id2 + t1
20:25
/43

24. Code Generation A compiler may generate Example: Advantages:
pure machine codes (machine dependent assembly language) directly, which is rare now ;
virtual machine code.
Example:
PASCAL  compiler  P-code  interpreter  execution
Speed is roughly 4 times slower than running directly generated machine codes.
Advantages:
simplify the job of a compiler;
decrease the size of the generated code: 1/3 for P-code ;
can be run easily on a variety of platforms
P-machine is an ideal general machine whose interpreter can be written easily;
divide and conquer;
recent example: JAVA and Byte-code.
20:25
/43

25. Code Generation Example
LDF R2, id3
MULF R2, R2, #60.0
LDF R1, id2
ADDF R1, R1, R2
STF id1, R1
t1 = id3 ＊ 60.0
id1 = id2 + t1
20:25
/43

26. Practical Considerations (1/2)
Preprocessing phase:
macro substitution:
#define MAXC 10
rational preprocessing: add new features for old languages.
BASIC
C  C ++
compiler directives:
#include <stdio.h>
non-standard language extensions.
adding parallel primitives
20:25
/43

27. Practical Considerations (2/2)
Passes of compiling
First pass reads the text file once.
May need to read the text one more time for any forward addressed objects, i.e., anything that is used before its declaration.
Example: C language
goto error_handling;
· · ·
error_handling:
20:25
/43

28. Reduce Number of Passes
Each pass takes I/O time.
Back-patching : leave a blank slot for missing information, and fill in the empty slot when the information becomes available.
Example: C language
when a label is used
if it is not defined before, save a trace into the to-be-processed table
label name corresponds to LABEL TABLE[i]
code generated: GOTO LABEL TABLE[i]
when a label is defined
check known labels for redefined labels
if it is not used before, save a trace into the to-be-processed table
if it is used before, then find its trace and fill the current address into the trace
Time and space trade-off !
20:25
/43