kumasento / dfe-mlir

MLIR dialect for data-flow engine design

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

dfe-mlir

MLIR dialect for data-flow engine design

Core Features

  • A maxj dialect that can concisely represent most data-flow designs achievable by MaxJ (e.g., multi-kernel and LMem designs). We can perform optimization on it, and designs described by it can be finally translated to valid MaxJ code.

Install

Please make sure your GCC and CMake satisfies the latest requirements from LLVM.

git clone --recursive https://github.com/kumasento/dfe-mlir

# build llvm
cd dfe-mlir/lib/llvm-project
mkdir build
cd build
cmake \
  -DLLVM_ENABLE_PROJECTS="clang;clang-tools-extra;lld;mlir" \
  -DLLVM_TARGETS_TO_BUILD="X86" \
  -DLLVM_ENABLE_ASSERTIONS=ON \
  -DLLVM_BUILD_LLVM_DYLIB=ON \
  -DLLVM_INCLUDE_TESTS=OFF -G "Ninja" ../llvm
ninja -j$(nproc)

# build dfe-mlir
cd ../../../
mkdir build
cd build
cmake -G "Ninja" ..
ninja

Test

The maxj Dialect

About

MLIR dialect for data-flow engine design

License:MIT License


Languages

Language:C++ 68.5%Language:MLIR 20.4%Language:CMake 8.5%Language:Python 2.6%