Filia is a project to build multi-language program verification and bug finding capabilities using the MLIR compiler infrastructure. MLIR is a general highly-extensible program intermediate representation suited for representing code writen in a variety of languages.

Filia currently includes: (1) py2mlir -- a Python to MLIR translation tool; (2) js2mlir -- a Javascript to MLIR translator; (3) script-opt-- an optimization tool that supports the Python and Javascript dialects.

js2mlir is self-contained and independent of the other components. See js2mlir/README.md for instructions on building and using js2mlir.

Filia requires a build of LLVM including MLIR and the MLIR Python bindings. We include a submodule that links to the version of LLVM used for testing Filia.

Filia requires a build of MLIR with the Python bindings. As these are not yet shipped with system versions of LLVM, we recommend that you clone the LLVM submodule and build this version. We also provide a script to configure LLVM with MLIR enabled. To use it, you need to first ensure that cmake, ninja and python with the PIP package manager are installed. Once these are installed, run the following commands from the Filia root directory:

source ./venv/bin/activate
pip install -r deps/llvm-project/mlir/python/requirements.txt
./deps/configure-llvm.sh
cmake --build deps/llvm-build --target install

This will install LLVM into the directory deps/llvm-install.

Once LLVM is installed, you can configure and build Filia with the following commands:

sudo cmake -S . -B build/debug -DMLIR_DIR=deps/llvm-install/lib/cmake/mlir
sudo cmake --build build/debug --target install

Fiia is still in active development and only supports a small subset of Python to see it working on an MD5 example, you can run the following commands:

./build/debug/py2mlir/py2mlir.py ./example/md5_example.py > md5.mlir

This MLIR file can then be optimized by script-opt. For example,

./dist/debug/bin/script-opt --python-load-store md5.mlir > md5-opt.mlir