KarolaScript on LLVM (project in progress)

The initial idea was to make "tree-walking code generation"; so source code goes through lexer and parser and finally ends up in AST.

That AST will be walked and LLVM IR would be generated during walk. Look in middleware/llvm-gen for these files.

.ks file --> AST --> LLVM IR --> assembly
          ^       ^           ^
        parser  lowering   backend (llc)

While working on that I found about something called MLIR.

The framework built for language hackers which can be used for sequential lowering.

It will help us when writing language specific optimizations.

Usefull resorces while on LLVM and MLIR journey:

• https://github.com/j2kun/mlir-tutorial
• https://mlir.llvm.org/docs/Tutorials/Toy/Ch-1/
• https://medium.com/sniper-ai/mlir-tutorial-create-your-custom-dialect-lowering-to-llvm-ir-dialect-system-1-1f125a6a3008
• https://www.youtube.com/watch?v=Ij4LswX1tZU&list=PLlONLmJCfHTo9WYfsoQvwjsa5ZB6hjOG5&index=1
• https://llvm.org/docs/TableGen/ProgRef.html
• https://mlir.llvm.org/docs/DefiningDialects/Operations/
• https://mlir.llvm.org/docs/PassManagement/#operation-pass-static-filtering-by-op-type
• Interesting Passes and Dialects:
  • https://mlir.llvm.org/docs/Dialects/MathOps/
  • https://mlir.llvm.org/docs/Dialects/ArithOps/
  • https://mlir.llvm.org/docs/Dialects/Linalg/
  • https://mlir.llvm.org/docs/Dialects/ControlFlowDialect/

---

Notes to self:

• from test.ll (IR code) to test.bc (Machine-level bitcode) use llvm-as. llvm-dis can be used for dissasembling.

To get the final executable from .ll:

1. Compile LLVM IR to bitcode:

llvm-as yourfile.ll -o yourfile.bc

2. Compile bitcode to native assembly:

llc yourfile.bc -o yourfile.s

3. Link the assembly to an executable:

clang yourfile.s -o yourfile