Parsing LLVM IR code (with debug symbols) to map it back to the original source

Question

I'm thinking about building a tool to help me visualise the generated LLVM-IR code for each instruction/function on my original source file. Something like this but for LLVM-IR.

The steps to build such tool so far seem to be:

• Start by with LLVM-IR AST builder.
• Parse generated IR code.
• On caret position get AST element.
• Read the element scope, line, column and file and signal it on the original source file.

Is this the correct way to approach it? Am I trivialising it too much?

Answer 1

I think your approach is quite correct. The UI part will probably be quite long to implement so I'll focus on the llvm part.

Let's say you start from a input file containing your LLVM-IR.

Step 1 process module:
Read file content to a string. Then Build a module from it, and process it to get the debug info:

llvm::MemoryBuffer* buf = llvm::MemoryBuffer::getMemBuffer(llvm::StringRef(fileContent)).release();
llvm::SMDiagnostic diag;
llvm::Module* module = llvm::parseIR(buf->getMemBufferRef(), diag, *context).release();
llvm::DebugInfoFinder* dif = new llvm::DebugInfoFinder();
dif->processModule(*module);

Step 2 iterate on instructions:
Once done with that, you can simply iterate on function and blocks and instructions:

// pseudo code for loops (real code is a bit long)
foreach(llvm::Function f in module.functions) 
{
   foreach(llvm::BasicBlock b in f.BasicBlockList)
   {
      foreach(llvm::Instruction inst in b.InstList) 
      {
         llvm::DebugLoc dl = inst.getDebugLoc();
         unsigned line = dl->getLine();
         // accordingly populate some dictionary between your instructions and source code
      }
   }
}

Step 3 update your UI
This is another story...