LLVM - How AST can be transformed to IR

Question

I know that an AST generated by the parser is used to generate IR in the frontend.

I am wondering how AST to be parsed and then transformed to IR (prob assembly or bitcode),

AST is a tree, what are the steps involved in the transformation from AST to IR.

Lodicule

Answer 1

Emitting LLVM IR from Clang ASTs happens in Clang's code gen stage. The code for this stage lives in lib/CodeGen/ (relative to Clang's source root). There's no need to parse the AST since Clang has the AST in an in-memory data structure. Code generation is essentially a recursive walk of the AST that emits IR into a Module. If there's any specific step of it that interests you, the best way to examine it would be to look in the code.

Sober

Thanks Eli. Would you mind to explain a bit more about what happened during the "walking of the AST that emits IR into a Module" ? – Lodicule

I mean, how AST can be transformed to IR ? Is that a mapping file or something else ? – Lodicule

@Lodicule explaining the process in detail will take a lot of time and space here, you'd better off checking the source by yourself – Incantatory

@MarcoA., thanks for your reply. Would it be possible to give a very brief explanation so that I can have a general sense before digging into the code ? – Lodicule

@Sam: the link given by mishr is an excellent simple example of emitting LLVM IR from a simple AST. Simple because the input language is simple. The real Clang AST represent languages that are anything but simple (C++), so walking them is very complex as well. – Sober

Thanks for the feedback. I will go through the link. – Lodicule

Here's an update to @shrm 's link: llvm.org/docs/tutorial/LangImpl03.html. As a starting point, the tutorial first provides a simple AST implementation. It then explains how a unique codegen() function can be written for each type of AST node you have. Then, simply walk the AST (calling the codegen() functions along the way) to generate the LLVM. – Doronicum

The problem I'm facing is to find where the codeGen process starts in clang. Can someone please give me some pointers? Thanks – Shelves