Compilation 2012 The What and Why of Compilers

Transcription

1 Compilation 2012 The What and Why of Compilers Jan Midtgaard Michael I. Schwartzbach Aarhus University

2 What is a Compiler? A program that: tralates from one programming language to another preserves the semantics (in some see) a compiler also implicitly defines a semantics but it is hard to reason about programs in terms of a compiler-d semantics Each language must have at least one compiler (or erpreter) It is often the case that: the source language is high-level the target language is low-level 2

3 New Programming Languages

4 Famous Programming Languages 1952 Autocoder 1956 IPL 1954 Fortran 1958 Lisp 1960 Algol 1960 Cobol 1964 APL 1964 PL/ BCPL 1967 Simula Snobol Algol Pascal 1970 SQL 1971 Forth 1972 Prolog 1972 Smalltalk 1973 C 1973 ML 1975 Scheme 1979 Modula Ada 1981 C SML 1985 Eiffel 1987 Caml 1987 Perl 1988 Oberon 1990 Haskell 1991 Python 1994 Dylan 1994 Java 1996 OCaml 1997 JavaScript 1997 Ruby 1998 Erlang 2000 C# 2001 AspectJ 2006 C 4

5 Famous Programming Languages 1952 Autocoder 1956 IPL 1954 Fortran 1958 Lisp 1960 Algol 1960 Cobol 1964 APL 1964 PL/ BCPL 1967 Simula Snobol Algol Pascal 1970 SQL 1971 Forth 1972 Prolog 1972 Smalltalk 1973 C 1973 ML 1975 Scheme 1979 Modula Ada 1981 C SML 1985 Eiffel 1987 Caml 1987 Perl 1988 Oberon 1990 Haskell 1991 Python 1994 Dylan 1994 Java 1996 OCaml 1997 JavaScript 1997 Ruby 1998 Erlang 2000 C# 2001 AspectJ 2006 C 5

6 Domain-Specific Languages Extend software design Are concrete artifacts that permit better: representation optimization analysis Replace low-level programs and libraries Exist in te of thousands Require full-scale compiler technology 6

7 Some Interesting DSL Compilers LaTeX document descriptio PostScript Esterel reactive systems hardware desig WS-BPEL flow algebras Web services Nyquist musical compositio sounds SQL queries evaluation pla 7

8 The FORTRAN Compiler Implemented in 1957 The world's first compiler Motivated by the economics of programming Had to overcome deep skepticism Focused on efficiency of the generated code Pioneered many concepts and techniques Revolutionized computer programming 8

9 How Good Are Compilers? /* naive */ for (i = 0; i < N; i++){ a[i] = a[i] * 2000; a[i] = a[i] / 10000; } /* expert */ b = a; for (i = 0; i < N; i++){ *b = *b * 2000; *b = *b / 10000; b++; } 9

10 Better Than We Are! /* naive */ for (i = 0; i < N; i++){ a[i] = a[i] * 2000; a[i] = a[i] / 10000; } /* expert */ b = a; for (i = 0; i < N; i++){ *b = *b * 2000; *b = *b / 10000; b++; } loop level sparc mips alpha naive naive -O naive -O expert expert -O expert -O

11 Better Than We Are! /* naive */ for (i = 0; i < N; i++){ a[i] = a[i] * 2000; a[i] = a[i] / 10000; } /* expert */ b = a; for (i = 0; i < N; i++){ *b = *b * 2000; *b = *b / 10000; b++; } loop level sparc mips alpha naive naive -O naive -O expert expert -O expert -O

12 Phases of Modern Compilers preprocessing scanning parsing desugaring weeding environments linking type checking static analysis resource allocation code generation code optimization code emission assembly 12

13 Compiler Architecure Compiler phases suggest a modular design: one phase = one module Many phases can be generated automatically Tools range from industrial to experimental Compilers divide o frontends and backends Both may be retargeted 13

14 A Tiny Java Method public void prnumber(, ) throws Exception { = ""; if ( == 0) { = "0"; } else { while ( > 0) { = ( % ) + ; = / ; } } System.out.pr(+"\n"); return; } 14

15 Stream of Toke keyword: public keyword: void identifier: prnumber symbol: ( keyword: identifier: symbol:, keyword: identifier: symbol: ) keyword: throws identifier: Exception symbol: { identifier: identifier: symbol: = ant: "" symbol: ; keyword: if symbol: ( identifier: symbol: == ant: 0 symbol: ) symbol: { identifier: symbol: = ant: "0" symbol: ; symbol: } keyword: else symbol: { keyword: while symbol: ( identifier: symbol: > ant: 0 symbol: ) symbol: { identifier: symbol: = symbol: ( identifier: symbol: % identifier: symbol: ) symbol: + identifier: symbol: ; identifier: symbol: = identifier: symbol: / identifier: symbol: ; symbol: }... 15

16 Abstract Syntax Tree method prnumber args type void sequence decl if System.out..pr type "" == while + name 0 "0" > sequence \n 0 + / % 16

17 Name and Type Linking method prnumber args type void sequence decl if System.out..pr type "" == while + name 0 "0" > sequence \n 0 + / % 17

18 18 Type Checking method prnumber type void args sequence decl type name "" if == 0 "0" while > 0 sequence concat % / System.out..pr concat \n boolean boolean

19 Type Checking Traformatio method prnumber args type void sequence decl if System.out..pr == type "" boolean while concat name 0 "0" > boolean sequence \n 0 concat / Operator + is traformed o concat % 19

20 20 Resource Allocation method prnumber type void args sequence decl type name "" if == 0 "0" while > 0 sequence concat % / System.out..pr concat+ \n boolean boolean 1 2 3

21 Bytecode Generation.method public prnumber(ii)v.throws java/lang/exception.limit stack 4.limit locals 4 ldc "" dup astore_3 pop 64 itructio iload_1 i_0 if_icmpeq true_2... stop_1: aload_3 invokestatic java/lang//valueof(ljava/lang/object;)ljava/lang/; ldc "\012" invokevirtual java/lang//concat(ljava/lang/;)ljava/lang/; getstatic java/lang/system/out Ljava/io/PrStream; swap invokevirtual java/io/prstream/pr(ljava/lang/object;)v return.end method 21

22 Bytecode Optimization.method public prnumber(ii)v.throws java/lang/exception.limit stack 4.limit locals 4 ldc "" astore_3 iload_1 i_0 40 itructio if_icmpne start_4 ldc "0" astore_3... stop_1: aload_3 invokestatic java/lang//valueof(ljava/lang/object;)ljava/lang/; ldc "\012" invokevirtual java/lang//concat(ljava/lang/;)ljava/lang/; getstatic java/lang/system/out Ljava/io/PrStream; swap invokevirtual java/io/prstream/pr(ljava/lang/object;)v return.end method 22

23 Code Assembly prnumber.class: cafe babe d a f 6c61 6e67 2f49 6e a f 6c61 6e67 2f4f 626a c69 6e69 743e c b c 6a f6c 616e 672f e67 3b29 4c6a f 6c61 6e67 2f e 673b 0c e0a 001f f6e f a53 6f c c00 0c a f6c 616e 672f e e 756c 6c a e f e 672e 6a01 000b e 744e 756d f 496e f e 670c f f e 670a a00 1f c6a f 6c61 6e67 2f e 673b c b d a b700 14b d c e 571b 039f a b 121b 594e 57a b03 a a abb 001f 591b 1c70 b700 0ab6 001d 2d59 c600 06a b6 001a 594e 571b 1c6c 593c 57a7 ffcf b1b e

24 Virtual Machines and Compilers Bytecode is just an ermediate representation of the program The VM is part of the compilation process But it is useful to separate the compilation process o (front-end) compiler and VM N languages and M platforms require only N (front-end) compilers and M virtual machines, rather than N*M monolithic compilers Enables adaptive optimization techniques VMs are quite complex pieces of software, too Have their own sophisticated architecture Register allocation, memory management, adaptive optimization, But in this course we will not say much about how VMs work

25 Bootstrapping Compilers (1/4) We are given: a machine language M a programming language L We need a compiler: source (this is called a T-diagram) L M M target implementation The direct approach is hard and difficult 25

26 Bootstrapping Compilers (2/4) We define: L is a simple subset of L M is naive and inefficient M code It is easy to implement: L M M And in parallel: L M L 26

27 Bootstrapping Compilers (3/4) Combining the two compilers we get: L M L M input program L L M M output program M A final combination gives us what we want: L M L M L L M M M 27

28 Bootstrapping Compilers (4/4) We now want a compiler for another language K We first program a compiler: K M L The new compiler is then obtained as: K M K M L L M M And so on... M 28