xDSL is a Python-native compiler framework built around SSA-based intermediate representations (IRs). Users of xDSL build a compiler by assembling predefined domain-specific IRs and, optionally, defining their own custom IRs. xDSL uses multi-level IRs, meaning that during the compilation process, a program will be lowered through several of these IRs. This allows the implementation of abstraction-specific optimization passes, similar to the structure of common DSL compilers (such as Devito, Psyclone, and Firedrake). To simplify the writing of these passes, xDSL uses a uniform data structure based on SSA, basic blocks, and regions, which additionally enables the writing of generic passes.
The design of xDSL is influenced by MLIR, a compiler framework developed in C++, that is part of the LLVM project. An inherent advantage of a close design is the easy interaction between the two frameworks, making it possible to translate abstractions and programs back and forth. This results in one big SSA-based abstraction ecosystem that can be worked with through Python, making analysis through simple scripting languages possible. Additionally, xDSL can leverage MLIR's code generation and low-level optimization capabilities.
To use xDSL as part of a larger project for developing your own compiler, just install xDSL via pip:
pip install xdslNote: This version of xDSL is validated against a specific MLIR version,
interoperability with other versions may result in problems. The supported
MLIR version is commit d401987fe349a87c53fe25829215b080b70c0c1a.
xDSL has a number of subprojects, some of which require extra dependencies. In order to keep the set of dependencies ot a minimum, these extra dependencies have to be specified explicitly. To install these, use:
pip install xdsl[gui,jax,riscv,wgpu,onnx]To install the testing/development dependencies, use:
pip install xdsl[dev]These may be useful for projects wanting to replicate the xDSL testing setup.
Check out the dedicated Getting Started guide for a comprehensive tutorial.
To get familiar with xDSL, we recommend starting with our Jupyter notebooks. The notebooks consist of examples and documentation concerning the core xDSL data structures and the xDSL's Python-embedded abstraction definition language, as well as examples of implementing custom compilers, like a database compiler. There also exists a small documentation showing how to connect xDSL with MLIR for users interested in that use case.
We provide a Makefile containing a lot of common tasks, which might provide an overview of common actions.
To contribute to the development of xDSL follow the subsequent steps.
We use uv for dependency management of xDSL.
Getting started documentation can be found here,
and is also printed by the make uv-installed and make venv targets if it
is not already installed on your system.
git clone https://github.com/xdslproject/xdsl.git
cd xdsl
# set up the venv and install everything
make venvTo make a custom mlir-opt available in the virtual environment, set the XDSL_MLIR_OPT_PATH variable when running make venv, like so:
XDSL_MLIR_OPT_PATH=/PATH/TO/LLVM/BUILD/bin/mlir-opt make venvFor some systems and workflows, changing to a new dependency management system
may be inconvenient, impractical, or impossible. If this is the case for you,
xDSL can still be installed using pip.
To create the required virtual environment (the equivalent of make venv):
python -m venv venv
source venv/bin/activate
pip install -e ".[dev]"The following commands can then be run using this virtual environment rather
than uv by running source venv/bin/activate when starting a new shell, then
eliding the uv run prefix from the commands. For example, to run the commands
in the following testing section:
source venv/bin/activate
pytest
lit tests/filecheckThe xDSL project uses pytest unit tests and LLVM-style filecheck tests. They can be executed from the root directory:
# Executes pytests which are located in tests/
uv run pytest
# Executes filecheck tests
uv run lit tests/filecheck
# run all tests using makefile
make testsAll python code used in xDSL uses ruff to format the code in a uniform manner.
To automate the formatting, we use pre-commit hooks from the pre-commit package.
# Install the pre-commit on your `.git` folder
make precommit-install
# to run the hooks:
make precommit
# alternatively, run ruff directly:
uv run ruff formatFurthermore, all python code must run through pyright
without errors. Pyright can be run on all staged files through the
makefile using make pyright.
Important
xDSL currently relies on an experimental feature of Pyright called TypeForm TypeForm is in discussion and will likely land in some future version of Python.
For xDSL to type check correctly using Pyright, please add this to your pyproject.toml:
[tool.pyright]
enableExperimentalFeatures = trueYou can also join the discussion at our Zulip chat room, kindly supported by community hosting from Zulip.