Clemson University -- CPSC 231 -- Fall 2009 Language translation approaches - compilation vs. interpretation * compilation diagram .---------. .------------------. step 1: compile | program | --> compiler --> | compiled program | `---------' `------------------' . . . . . .-------. v .--------. step 2: run | input | --> compiled program --> | output | `-------' `--------' * compilation is translation from one language to another, where the translated form is typically easier to execute; a pure compiler produces language that will be directly executed by hardware * compilation allows one translation and then multiple executions of the executable file (sometimes called a binary file, or load module); thus a fairly large amount of time can be spent by the compiler doing analysis and optimization once, in order to produce an executable that runs quickly each time it is run * a compiled program typically runs fast but is harder to debug * compiler example: gcc * interpretation diagram .---------. single step | program | ---. `---------' v .--------. interpreter --> | output | .-------. ^ `--------' | input | -----' `-------' * interpretation skips the intermediate step of producing a form of the program in another language and combines translation and execution * interpretation starts from the source code each time you want to run the program; it performs the same analysis as a compiler but on a source-line-by-source-line basis; a pure interpreter keeps no results from this analysis even when encountering the same source line repeatedly within the body of a loop (this means an interpreted program will run faster if you make all the variable and function names only one or two characters in length and remove all the comments -- but I don't recommend doing this!) * an interpreted program typically runs slow but is easier to debug because of better run-time error diagnostics * interpreted languages easily support dynamic typing and dynamic scoping of variables * interpreter examples: shells, m4 or python on the command line; also, formatted I/O (e.g., printf) relies on interpretation * hybrid approach diagram .---------. .-----------. step 1: | program | --> compiler --> | byte code | `---------' `-----------' . . . . . . . . . . . . . . . v .-----------. step 2: | byte code |---. `-----------' v .--------. JVM --> | output | .-------. ^ `--------' | input | ------' `-------' * Java compiler and JVM interpreter - a hybrid translation model - "javac" produces byte code, which is easy to interpret - "java" interprets byte code * provides for portability of byte code files across numerous systems * Perl also has a hybrid translation model * other hybrid translation models include just-in-time (JIT) compilers, which compile functions/procedures at run-time, on the first call * terminology - source code that needs to be compiled is typically called a "program" while source code that is interpreted may be called a "script" (but may also be called a "program" also) major translators in compilation model 1. language preprocessor - textual substitution and conditional compilation (direct execution of special statements) 2. compiler - lexical analysis, parsing, code generation, optimization 3. macro processor - textual substitution and conditional assembly 4. assembler - translate symbols into addresses and machine code 5. linker - external symbol resolution plus relocation, produces executable 6. loader - relocation according to load address, produces memory image compile steps ------------- language preprocessor compiler assembler linker (cpp) (ccom) (as) (ld) .------. .--------. .--------. .------. .-------------. |source|------->|expanded|----->|assembly|------>|object|--->| executable | | code | | source | |language| | code | |(load module)| | (.c) | | code | | (.s) | | (.o) | | (a.out) | `------' `--------' | (.asm) | |(.obj)| | (.exe) | macro `--------' `------' `-------------' expansion and compile ^ assembly link ^ conditional time | time time | compilation | / | .-------. / / |library|---- macro processor / |routine| (m4) / `-------' --------------- / |assembly source|------------ static linking |w/ macros (.m) | `--------------- macro expansion and conditional assembly load and run steps ------------------ command interpreter (shell) loader fetch/decode/execute on CPU .-------------. .................................. search for | executable |----->. memory (... machine lang. ...) . file name |(load module)| . image (...insts. and data...) . | (a.out) | .................................. | (.exe) | ^ ^ ^ `-------------' . | | . | | . | | load-time linking . / | run-time linking (early Windows) . / | (most systems) . / | .---------------. . / ................... | library files |--------------' . shared object . |(Microsoft DLL)| . (.so) . `--------------- ................... dynamic linking translators - language preprocessor, e.g, for C * special syntax for preprocessor statements, e.g., #include * macro facility, #define - trivially used for constant substitution * conditional compilation, #ifdef - used for versioning #ifdef VERBOSE printf( "value of a is %d\n", a ); #endif where "#define VERBOSE" is included in the program source or where you compile with "gcc -DVERBOSE" - compiler * lexical analysis - extracting lexical items ("tokens") from the input * syntactic analysis - parsing statements according to the grammar rules the of language, generates a parse tree * semantic analysis - determining the meaning of operations according to the datatypes of the variables in the parse tree, may involve adding conversion operators to the parse tree * intermediate code generation * machine-independent optimizations, e.g., loop transformations * machine-specific code generation and register allocation * machine-dependent optimizations, e.g., branch delay slot scheduling consider the statement a = b + 2*c; in the following code float a,b; extern float c; ... a = b + 2*c; ... lexical analysis extracts eight tokens and assigns symbolic identifiers to entries in the symbol table `a' `=' `b' `+' `2' `*' `c' `;' symtab[0] `=' symtab[1] `+' `2' `*' symtab[2] `;' syntactic analysis builds a parse tree = / \ symtab[0] + / \ symtab[1] * / \ `2' symtab[2] semantic analysis determines meaning =:float / \ symtab[0]:float +:float / \ symtab[1]:float *:float / \ convert_to_float symtab[2]:float | `2' intermediate code generation yields something like convert_to_float( 2 , temp_float_0 ) multiply_float( temp_float_0 , symtab[2] , temp_float_1 ) add_float( symtab[1] , temp_float_1 , temp_float_2 ) store_float( temp_float_2 , symtab[0] ) machine-independent optimization goes ahead and either does the conversion at compile time or strength reduces the multiply by 2 to an add add_float( symtab[2] , symtab[2] , temp_float_1 ) add_float( symtab[1] , temp_float_1 , temp_float_2 ) store_float( temp_float_2 , symtab[0] ) from this registers would be assigned and SPARC code would be generated (including storage allocation and addressing for variables) .global c ... save %sp, -128, %sp ! enough room for locals a and b ... sethi %hi(c), %o0 ! linker has to fill in %hi(c) bits ld [ %o0 + %lo(c) ], %f2 ! linker has to fill in %lo(c) bits fadds %f2, %f2, %f4 ld [ %fp + b_offset ], %f6 fadds %f6, %f4, %f8 st %f8, [ %fp + a_offset ] example of optimizations typedef struct node{ int data; struct node *left_child; struct node *right_child; } NODE; int binary_search( NODE *ptr, int item ){ if( ptr == NULL ) return( 0 ); else if( ptr->data == item ) return( 1 ); else if( item < ptr->data ) return( binary_search( ptr->left_child, item )); else return( binary_search( ptr->right_child, item )); } binary_search: /* produced using gcc -S -O2 */ .LL12: cmp %o0, 0 /* leaves 0 for NULL in %o0 as retval */ be .LL1 nop ld [%o0], %g1 cmp %g1, %o1 be,a .LL14 /* annulling branch optimization */ mov 1, %o0 /* (during delay slot scheduling) */ ble,a .LL12 /* tail call elimination (branch to */ ld [%o0+8], %o0 /* top instead of recursive call) */ b .LL12 ld [%o0+4], %o0 .LL14: .LL1: retl /* leaf procedure optimization */ nop - macro processor * simple abstraction through textual substitution ("open" subroutines) * provides either keyword or positional parameter substitution * extends inst. set by synthesizing instructions using macro definitions (similar to what is already done by SPARC assembler, cf. "set") * cost occurs at assembly time of expanding macro definition, not at run time of procedure call, register save/restore, and procedure return * conditional assembly is same idea as #ifdef facility of C preprocessor comparison with run-time functions macro function ----- -------- invocation in-line substitution run-time call and return parameters untyped typed evaluated at each evaluated once at time appearance of call trade-offs fast but one copy of more overhead per call but code at each call site only one copy of code - assembler * translates program written in assembly language to binary machine code resolving local symbolic addresses; typically this is 1-to-1 translation * forward references generally require 2-pass assemblers pass 1: find symbolic labels and assign them addresses run location counter (virtual instruction pointer) determine instruction size record addresses in symbol table pass 2: use symbol table information to construct instructions symbolic -> binary * alternative to 2-pass approach is 1-pass with fixup (i.e., backpatching) * other assembler facilities include data layout directives (pseudo-ops) - linker * separate assembly or compilation means the assembler does not know all the addresses, thus the assembler produces only partially-resolved object files * linker combines separate object files into a single executable - layout pieces of code & data (storage allocation based on sizes) - resolve external references - perform relocation of absolute addresses * two pass: 1: assign code and data to memory addresses and build symbol table from public symbols 2: use table to resolve external addresses and produce load module * object module file format (this is early UNIX; ELF is more complex) - header (includes sizes of text, data, and bss sections) - text section (read only) - data section (read/write) - relocation/external symbol entries for text section - relocation/external symbol entries for data section - symbol table - string table (symbol table entries index into string table) - command interpreter (shell) - a program that reads command lines from the keyboard (or from a script file) and either directly executes the command or searches for an executable file having that command name and then loads and branches to that loaded program - loader * bring a program into memory in preparation for execution * read file header to find size of pieces * allocate memory area(s) * read instructions and data from file into memory * relocation - adjusting absolute addresses relative to load point * jump to startup code binding times The assembler, linker, and loader are all programs taking input files and producing output. Decisions and translations made by these programs are said to be done at "assembly time", at "link time", and at "run time", respectively. Actual execution (i.e., instruction interpretation by the hardware such as performing adds, branches, etc.) takes place at "run time". During execution, you can also talk of things happening at specific times, such as register saving at procedure call time. Dynamic linking is an example of a late decision, or "late binding". It is the linking of separate procedures at either load time or run time, and it typically requires that the normal (static) linker include a simple table that names the needed routines (for load-time linking) or include simple "stub" routines that find and link to the shared library routines on their first calls (for run-time linking). Another form of delayed binding is "just-in-time" (JIT). This is used in several Java compilers, where methods are not compiled until the first call. Many storage allocation decisions are made at each step. For example, offsets are assigned to labels at assembly time, under the assumption that any absolute addresses will be updated by the linker and loader later. (When we later study virtual memory, we will see that it is also an example of late binding - specifically one where physical memory allocation decisions that might be made by a traditional loader are instead deferred to run time and made by the operating system.) other programming tools / components of a program development environment editors (e.g., vi, emacs) beautifiers (e.g., indent) project control (e.g., make) version control (e.g., sccs) GUI toolkit (e.g., widget library) test coverage (e.g., gcov) debuggers (e.g., gdb, dbx, ddd) debugging tools (e.g., Purify) reading or writing beyond the bounds of an array reading or writing freed memory freeing memory multiple times reading uninitialized memory reading or writing through null pointers overflowing the stack by recursive function calls reading or writing memory addresses on which a watch-point has been set portability advisors (e.g., lint) style checkers (e.g., CodeCheck) exceeding a given input line length exceeding a given nesting depth of if-else stmts. not aligning open and close curly braces (Horstmann) performance profilers (e.g., gprof)