ABSTRACT
Early techniques in code generation centered around interpretive approaches where code is produced for a virtual machine and then expanded into real machine instructions. The interpretive approach suffers from the drawback of having to change the code generator for each machine. The idea of code generation by tree parsing replaced the strategy of virtual machine interpretation. The intermediate representation (IR) of the source program is in the form of a tree and the target machine instructions are represented as productions of a regular tree grammar augmented with semantic actions and costs. The code generator parses the input subject tree and, on each reduction, outputs target code. This is illustrated in Figure 15.1 for a subject tree generated by the grammar of Example 15.1.
