P4MLIR

WARNING: This is an experimental project in incubation, not ready for any serious use yet.

P4MLIR aims to explore leveraging MLIR in building P4 compilers.

P4MLIR is structured as both a standalone MLIR project, and a P4C extension. Once it's mature enough, the plan is to merge it into the P4C repo.

Developer guide

Build from source

Check out the code

git clone --recursive [email protected]:p4lang/p4mlir.git
cd p4mlir

If you forgot --recursive, additionally run:

git submodule update --init --recursive

Build MLIR from source

Install common LLVM/MLIR requirements following the instructions. Or use our script if you are on Ubuntu:

./build_tools/ubuntu_install_mlir_requirements.sh

Then build MLIR from source and install it:

./build_tools/build_mlir.sh

Sometimes this script will stop in the middle and report some failures. Re-running the script will continue the build process and usually go past the last failure. Eventually the build shall finish successfully.

Some MLIR regression tests have already been performed during the build process. If you want to run them again, do the following:

cd third_party/llvm-project/build
ninja check-mlir

IMPORTANT: P4C uses C++ RTTI and Exceptions. These are turned off by default in LLVM and therefore LLVM/MLIR prebuilt binaries that is likely available from your distribution will be not usable with P4MLIR. One needs to build LLVM/MLIR from source enabling RTTI (-DLLVM_ENABLE_RTTI=ON) and exceptions (-DLLVM_ENABLE_EH=ON).

Build P4C with P4MLIR extension from source

./build_tools/build_p4c_with_p4mlir_ext.sh

Testing

cd third_party/p4c/build
ninja check-p4mlir