Now, since the AST is ready with all the necessary information in its data structure, the next phase is to generate LLVM IR. LLVM APIs are used in this code generation. LLVM IR has a predefined format that is generated by the inbuilt APIs of LLVM.
In order to generate LLVM IR, a virtual CodeGen function is defined in each AST class (the AST classes were defined earlier in the AST section; these functions are additional to those classes) as follows:
- Open the
toy.cppfile as follows:$ vi toy.cpp
- In the
BaseASTclass defined earlier, append theCodegen()functions as follows:class BaseAST { … … virtual Value* Codegen() = 0; }; class NumericAST : public BaseAST { … … virtual Value* Codegen(); }; class VariableAST : public BaseAST { … … virtual Value* Codegen(); };This virtual
Codegen()function is included in every AST class we defined.This function returns an LLVM Value object, which represents Static Single Assignment (SSA) value in LLVM. A few more static variables are defined that will be used during Codegen.
- Declare the following static variables in global scope as follows:
static Module *Module_Ob; static IRBuilder<> Builder(getGlobalContext()); static std::map<std::string, Value*>Named_Values;
The Module_Ob module contains all the functions and variables in the code.
The Builder object helps to generate LLVM IR and keeps track of the current point in the program to insert LLVM instructions. The Builder object has functions to create new instructions.
The Named_Values map keeps track of all the values defined in the current scope like a symbol table. For our language, this map will contain function parameters.