Lecture Notes on Intermediate Representation 15-411: Compiler Design

In this lecture we discuss the “middle end” of the compiler. After the source has been parsed we obtain an abstract syntax tree, on which we carry out various static analyses to see if the program is well-formed. In the L2 language, this consists of checking that every finite control flow path ends in a return statement, that every variable is initialized before its use along every control flow path, and that break and continue statements occur only inside loops. In later languages, type-checking will be an important additional task. After we have constructed and checked the abstract syntax tree, we transform the program through several forms of intermediate representation on the way to abstract symbolic assembly and finally actual x86-64 assembly form. How many intermediate representations and their precise form depends on the context: the complexity and form of the language, to what extent the compiler is engineered to be retargetable to different machine architectures, and what kinds of optimizations are important for the implementation. Some of the most well-understood intermediate forms are intermediate representation trees (IR trees), static single-assignment form (SSA), quads and triples. Quads (that is, three-address instructions) and triples (two-address instructions) are closer to the back end of the compiler and you will probably want to use one of them, maybe both. In this lecture we focus on IR trees.