/mlir-tutorial

MLIR For Beginners tutorial

Primary LanguageC++

MLIR For Beginners

This is the code repository for a series of articles on the MLIR framework for building compilers.

Articles

  1. Build System (Getting Started)
  2. Running and Testing a Lowering
  3. Writing Our First Pass
  4. Using Tablegen for Passes
  5. Defining a New Dialect
  6. Using Traits
  7. Folders and Constant Propagation
  8. Verifiers
  9. Canonicalizers and Declarative Rewrite Patterns
  10. Dialect Conversion
  11. Lowering through LLVM
  12. A Global Optimization and Dataflow Analysis
  13. Defining Patterns with PDLL

Bazel build

Bazel is one of two supported build systems for this tutorial. The other is CMake. If you're unfamiliar with Bazel, you can read the tutorials at https://bazel.build/start. Familiarity with Bazel is not required to build or test, but it is required to follow the articles in the tutorial series and explained in the first article, Build System (Getting Started). The CMake build is maintained, but was added at article 10 (Dialect Conversion) and will not be explained in the articles.

Prerequisites

Install Bazelisk via instructions at https://github.com/bazelbuild/bazelisk#installation. This should create the bazel command on your system.

You should also have a modern C++ compiler on your system, either gcc or clang, which Bazel will detect.

Build and test

Run

bazel build ...:all
bazel test ...:all

CMake build

CMake is one of two supported build systems for this tutorial. The other is Bazel. If you're unfamiliar with CMake, you can read the tutorials at https://cmake.org/getting-started/. The CMake build is maintained, but was added at article 10 (Dialect Conversion) and will not be explained in the articles.

Prerequisites

Checking out the code

Checkout the tutorial including the LLVM dependency (submodules):

git clone --recurse-submodules https://github.com/j2kun/mlir-tutorial.git
cd mlir-tutorial

Building dependencies

Note: The following steps are suitable for macOs and use ninja as building system, they should not be hard to adapt for your environment.

Build LLVM/MLIR

#!/bin/sh

BUILD_SYSTEM=Ninja
BUILD_TAG=ninja
THIRDPARTY_LLVM_DIR=$PWD/externals/llvm-project
BUILD_DIR=$THIRDPARTY_LLVM_DIR/build
INSTALL_DIR=$THIRDPARTY_LLVM_DIR/install

mkdir -p $BUILD_DIR
mkdir -p $INSTALL_DIR

pushd $BUILD_DIR

cmake ../llvm -G $BUILD_SYSTEM \
      -DCMAKE_CXX_COMPILER="$(xcrun --find clang++)" \
      -DCMAKE_C_COMPILER="$(xcrun --find clang)" \
      -DCMAKE_INSTALL_PREFIX=$INSTALL_DIR \
      -DLLVM_LOCAL_RPATH=$INSTALL_DIR/lib \
      -DLLVM_PARALLEL_COMPILE_JOBS=7 \
      -DLLVM_PARALLEL_LINK_JOBS=1 \
      -DLLVM_BUILD_EXAMPLES=OFF \
      -DLLVM_INSTALL_UTILS=ON \
      -DCMAKE_OSX_ARCHITECTURES="$(uname -m)" \
      -DCMAKE_BUILD_TYPE=Release \
      -DLLVM_ENABLE_ASSERTIONS=ON \
      -DLLVM_CCACHE_BUILD=ON \
      -DCMAKE_EXPORT_COMPILE_COMMANDS=ON \
      -DLLVM_ENABLE_PROJECTS='mlir' \
      -DDEFAULT_SYSROOT="$(xcrun --show-sdk-path)" \
      -DCMAKE_OSX_SYSROOT="$(xcrun --show-sdk-path)"

cmake --build . --target check-mlir

popd

Build and test

#!/bin/sh

BUILD_SYSTEM="Ninja"
BUILD_DIR=./build-`echo ${BUILD_SYSTEM}| tr '[:upper:]' '[:lower:]'`

rm -rf $BUILD_DIR
mkdir $BUILD_DIR
pushd $BUILD_DIR

LLVM_BUILD_DIR=externals/llvm-project/build
cmake -G $BUILD_SYSTEM .. \
    -DLLVM_DIR="$LLVM_BUILD_DIR/lib/cmake/llvm" \
    -DMLIR_DIR="$LLVM_BUILD_DIR/lib/cmake/mlir" \
    -DBUILD_DEPS="ON" \
    -DBUILD_SHARED_LIBS="OFF" \
    -DCMAKE_BUILD_TYPE=Debug

popd

cmake --build $BUILD_DIR --target MLIRAffineFullUnrollPasses
cmake --build $BUILD_DIR --target MLIRMulToAddPasses
cmake --build $BUILD_DIR --target MLIRNoisyPasses
cmake --build $BUILD_DIR --target mlir-headers
cmake --build $BUILD_DIR --target mlir-doc
cmake --build $BUILD_DIR --target tutorial-opt
cmake --build $BUILD_DIR --target check-mlir-tutorial