FSharp_MLIR

FSharp MLIR is a compiler for the FSharp programming language that uses the MLIR infrastructure to lower the FSharp code to LLVM IR and then to machine code.

It is a research project that aims to explore the possibilities of using MLIR to lower a functional programming language to machine code.

The supported feature set is currently very limited, but simple test programs can be compiled and run using the FSharpCompilerApp.

Requirements

Platform Support

Currently only Linux development is supported. While other platforms may work, they are not tested and may require additional setup. If you are feeling adventurous, you can try to change the build system following the MLIR documentation.

If you are using Windows, you can use the Windows Subsystem for Linux (WSL) to build and run the project. We recommend using WSL2 with a recent Ubuntu distro or Debian if you want a more minimal experience.

Using Debian will probably require setup of more dependencies, but it should be possible to build the project. Check the output of the build system for missing dependencies and install them as needed.

Compiler Support

You will need a compiler that supports C++20 (no module support required) and CMake 3.20 or newer.

Since this project uses the LLVM project extensively we recommend using the Clang compiler to build the project since it is the most tested compiler with LLVM, and also the only one that we can guarantee will work.

This project was developed using Clang 18. Use the scripts below to install the required dependencies.

Building

Dependencies

FSharp MLIR requires the following dependencies to be installed on the system:

sudo bash -c "$(wget -O - https://apt.llvm.org/llvm.sh)" 
sudo apt install default-jdk 
sudo apt install ccache 
sudo apt install cmake 
sudo apt install build-essential 
sudo apt install lld 
sudo apt install ninja-build

Building the Project

Clone the project into a directory of your choice and build the project using the following commands:

git clone https://github.com/LaFoster00/FSharp_MLIR.git --recursive --progress --shallow-submodules 
cd FSharp_MLIR 
mkdir build 
cd build 
cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_COMPILER=/usr/bin/clang++-18 -DCMAKE_C_COMPILER=/usr/bin/clang-18
ninja FSharpCompilerApp

If you want to only build the another target use the following command:

ninja target_name

Possible target are:

ninja FSharpCompilerApp

The GTEST targets:

ninja FSharpCompilerTests

ninja FSharpGrammarTests

Some legacy test targets:

ninja FSharpGrammarTimedTest

Configuring the project the first time will take a while since it will build the LLVM project. It will also download a couple of additional dependencies such as boost which will also take some time.

Running the Compiler

To run the compiler use the following command:

./compiler/app/FSharpCompilerApp -emit=exe -o <path-to-exe-outpout> <path-to-fsharp-source-file>

Possible options are:

./compiler/app/FSharpCompilerApp --help
OVERVIEW: fsharp compiler

USAGE: FSharpCompilerApp [options] <input fsharp file>

OPTIONS:

FSharp Compiler Options:
Specific options for FSharp compiler

  --emit=<value> - Select the kind of output desired
    =st          -   output the ST dump
    =ast         -   output the AST dump
    =mlir        -   output the MLIR dump
    =mlir-typed  -   output the MLIR dump after type inference
    =mlir-arith  -   output the MLIR dump after arith lowering
    =mlir-func   -   output the MLIR dump after func lowering
    =mlir-buff   -   output the MLIR dump after bufferization
    =mlir-llvm   -   output the MLIR dump after llvm lowering
    =llvm        -   output the LLVM IR dump
    =jit         -   JIT the code and run it by executing the input script
    =exe         -   Emit an executable from the input script
  -o <filename>  - Specify the output filename for the executable
  --opt          - Enable optimizations
  -x=<value>     - Decided the kind of output desired
    =fsharp      -   load the input file as a FSharp source.
    =mlir        -   load the input file as an MLIR file

Generic Options:

  --help         - Display available options (--help-hidden for more)
  --help-list    - Display list of available options (--help-list-hidden for more)
  --version      - Display the version of this program