DISCLAIMER: This is not an officially-supported Google project. It is a sandbox for quick iteration and experimentation on projects related to the IREE project, MLIR, and LLVM.
This repository contains experimental work by the IREE team closely related to LLVM and MLIR, usually with the aim of upstreaming in some form. The main project is at https://github.com/google/iree.
As an experimental project, build greenness, documentation, and polish are likely to be minimal, as it instead prioritizes easy experimentation.
Licensed under the Apache license with LLVM Exceptions. See LICENSE for more information.
Export some useful environment variables (add them to your ~/.bashrc) and
mkdir the directories:
export LLVM_SOURCE_DIR=${HOME}/github/llvm-project && \
export LLVM_BUILD_DIR=${HOME}/github/builds/llvm && \
export LLVM_INSTALL_DIR=${HOME}/github/install/ && \
export IREE_LLVM_SANDBOX_SOURCE_DIR=${HOME}/github/iree_llvm_sandbox && \
export IREE_LLVM_SANDBOX_BUILD_DIR=${HOME}/github/builds/iree_llvm_sandbox && \
export NPCOMP_SOURCE_DIR=${HOME}/github/mlir-npcomp && \
export NPCOMP_BUILD_DIR=${HOME}/github/builds/npcomp
Follow the instructions for MLIR Python Bindings:
which python
python -m venv ~/.venv/mlirdev
source ~/.venv/mlirdev/bin/activate
python -m pip install --upgrade pip
python -m pip install -r ${LLVM_SOURCE_DIR}/mlir/lib/Bindings/Python/requirements.txt
Optionally, install pytorch nightly:
pip3 install --pre torch torchvision torchaudio -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html
Get LLVM, for instance:
git clone git@github.com:llvm/llvm-project.git ${LLVM_SOURCE_DIR}
Build and install LLVM + MLIR with python bindings (also see the mlir getting started doc):
(cd ${LLVM_SOURCE_DIR} && \
\
cmake -G Ninja llvm \
-Dpybind11_DIR=${HOME}/.venv/mlirdev/lib/python3.9/site-packages/pybind11/share/cmake/pybind11/ \
-DLLVM_ENABLE_PROJECTS="mlir" \
-DBUILD_SHARED_LIBS=ON \
-DLLVM_BUILD_LLVM_DYLIB=ON \
-DLLVM_BUILD_EXAMPLES=ON \
-DLLVM_TARGETS_TO_BUILD="X86" \
-DMLIR_INCLUDE_INTEGRATION_TESTS=ON \
-DCMAKE_BUILD_TYPE=Release \
-DMLIR_BINDINGS_PYTHON_ENABLED=ON \
-DPython3_EXECUTABLE=$(which python) \
-DLLVM_ENABLE_ASSERTIONS=ON \
-DCMAKE_INSTALL_PREFIX=${LLVM_INSTALL_DIR} \
-DLLVM_INCLUDE_UTILS=ON \
-DLLVM_INSTALL_UTILS=ON \
-B ${LLVM_BUILD_DIR} && \
\
cmake --build ${LLVM_BUILD_DIR} --target check-mlir; \
\
cmake --build ${LLVM_BUILD_DIR} --target install)
Verify the MLIR cmake has been properly installed:
find ${LLVM_INSTALL_DIR} -name MLIRConfig.cmake
This should print: ${LLVM_INSTALL_DIR}/lib/cmake/mlir/MLIRConfig.cmake
Get iree-llvm-sandbox:
git clone git@github.com:google/iree-llvm-sandbox.git ${IREE_LLVM_SANDBOX_SOURCE_DIR}
Build iree-llvm-sandbox:
(cd ${IREE_LLVM_SANDBOX_SOURCE_DIR} && \
\
cmake -GNinja \
-DMLIR_DIR=${LLVM_INSTALL_DIR}/lib/cmake/mlir \
-DCMAKE_BUILD_TYPE=Debug \
-B ${IREE_LLVM_SANDBOX_BUILD_DIR} && \
\
cmake --build ${IREE_LLVM_SANDBOX_BUILD_DIR} --target all)
Run a simple sanity check:
LD_LIBRARY_PATH=${IREE_LLVM_SANDBOX_BUILD_DIR}/runners/lib \
${IREE_LLVM_SANDBOX_BUILD_DIR}/runners/mlir-proto-opt \
${IREE_LLVM_SANDBOX_SOURCE_DIR}/runners/test/test_constant.mlir \
-linalg-comprehensive-bufferize-inplace
Set up you PYTHONPATH properly:
export PYTHONPATH=${PYTHONPATH}:$LLVM_INSTALL_DIR/python:${IREE_LLVM_SANDBOX_BUILD_DIR}:${IREE_LLVM_SANDBOX_BUILD_DIR}/runners/lib; \
export PYTHONPATH=${PYTHONPATH}:${NPCOMP_BUILD_DIR}:${NPCOMP_BUILD_DIR}/lib:${NPCOMP_BUILD_DIR}/python
Run a simple python sanity check:
python ${IREE_LLVM_SANDBOX_SOURCE_DIR}/runners/test/python/linalg_matmul.py
Optionally, get npcomp-mlir:
git clone git@github.com:llvm/mlir-npcomp.git ${NPCOMP_SOURCE_DIR}
Optionally build npcomp-mlir:
(cd ${NPCOMP_SOURCE_DIR} && \
\
cmake -GNinja \
-Dpybind11_DIR=${HOME}/.venv/mlirdev/lib/python3.9/site-packages/pybind11/share/cmake/pybind11/ \
-DMLIR_DIR=${LLVM_INSTALL_DIR}/lib/cmake/mlir \
-DCMAKE_BUILD_TYPE=Debug \
-DPYTHON_EXECUTABLE=$(which python) \
-DPython3_EXECUTABLE=$(which python) \
-DCMAKE_BUILD_TYPE=Debug \
-DNPCOMP_USE_SPLIT_DWARF=ON \
-DCMAKE_CXX_FLAGS_DEBUG=$DEBUG_FLAGS \
-DLLVM_ENABLE_WARNINGS=ON \
-DCMAKE_EXPORT_COMPILE_COMMANDS=TRUE \
-B ${NPCOMP_BUILD_DIR} && \
\
cmake --build ${NPCOMP_BUILD_DIR} --target all)
TODOs:
- hook up a lit test target.
Python tests come with a tool to perform as simple randomized search. The search is going to randomly instantiate a given op to some cocnrete dimensions and type variables and try to compile it using mlir.
The results are persisted in the output/ folder by default in a structure that
includes a name of the expert compiler, the name of the op and the
success/failure/timeout status code. The results contain the full program output
(including potential compilation errors) and an accompanying .sh file that can
be used to re-run the same configuration again.
To run the search with default settings:
search_cli=${IREE_LLVM_SANDBOX_SOURCE_DIR}/runners/test/python/search_cli.py
python3 $search_cli
To run with a different linalg op, use --op flag:
python3 $search_cli --op matvec
To specify the name of the expert compiler, use --expert (see experts.py for
all available expert definitions):
python3 $search_cli --expert ExpertCompiler1
To specify the possible types, use --types flag:
python3 $search_cli --types f32,f64
Alternatively, one can also force some variables to concrete values, while
others will ramain random using --assign:
python3 $search_cli --assign M=16 N=32 K=64
To specify range of possible values for dimensions, use --range flag (where
numbers correspond to arguments of the corresponding range function in
Python):
python3 $search_cli --range 128,256,8
The search can be run using multiple processes at once, via --par flag:
python3 $search_cli --par 72
Each process collects the fixed number of random samples, customized via
--samples flag:
python3 $search_cli --samples 100