编译_chapter07.pptx

Translation to Intermediate Code 邱锡鹏复旦大学计算机科学技术学院mailto:xpqiu@xpqiu@/~xpqiu/compiler/~/~xpqiu/compilerxpqiu/compilerReferences/class/cs143//class/cs143/class/cs143//Partial contents are copied from the slides of Alex AikenOutlineIR TreesTranslation into TreesWhy Intermediate Representations (IR)?Individual pieces of abstract syntax can be complicated thingssuch as array subscripts, procedure calls, and so onIndividual “real machine” instructions can also a complicated effect.Therefore, the IR should be have individual components that describe only extremely simple thing:a single fetch, store, add, move or jump.Intermediate Representations (IR)What makes a good IR ? easy to convert from the AST; easy to convert into the machine code; must be clear and simple; must support various machine-independent optimizing transformations;VS Intermediate languageA popular format for intermediate languages is three address code.C is used as an intermediate language by many programming languages.Sun’s Java bytecode and Microsofts Common Intermediate LanguageIntermediate CodeUsually two IRs:Low-level IRMachine independent(but closer to machine)High-level IRLanguage-independent(but closer to language)Several IRsSome modern compilers use several IRs ( e.g., k=3 in SML/NJ ) each IR in later phase is a little closer (to the machine code) than the previous phase.AST == IR1 == IR2 ... == IRk == machine codepros : make the compiler cleaner, simpler, and easier to maintaincons : multiple passes of code-traversal --- compilation may be slowHigh-level IRTree node structure very similar to the ASTContains high-level constructs common to many languagesExpression nodesStatement nodesExpression nodes for:Integers and program variablesBinary operations: e1 OP e2Arithmetic operationsLogic operationsComparisonsUnary operations: OP eArray accesses: e1[e2]High-level IRStatement nodes:Block statements (statement sequences): (s1, …, sN)Variable assignments: v = eArray assignments: e1[e2] = e3If-then-