构建说明

要求

• 可用的C和C++工具链（编译器、链接器）
• cmake
• make或ninja

1. 克隆Polygeist

git clone --recursive https://github.com/llvm/Polygeist
cd Polygeist

2. 安装LLVM、MLIR、Clang和Polygeist

选项1：使用预构建的LLVM、MLIR和Clang

Polygeist可以通过提供预构建的MLIR和Clang工具链的路径来构建。

1. 构建LLVM、MLIR和Clang：

mkdir llvm-project/build
cd llvm-project/build
cmake -G Ninja ../llvm \
  -DLLVM_ENABLE_PROJECTS="mlir;clang" \
  -DLLVM_TARGETS_TO_BUILD="host" \
  -DLLVM_ENABLE_ASSERTIONS=ON \
  -DCMAKE_BUILD_TYPE=DEBUG
ninja
ninja check-mlir

要启用CUDA编译，请添加-DMLIR_ENABLE_CUDA_RUNNER=1并从cmake参数中移除-DLLVM_TARGETS_TO_BUILD="host"。（您可能需要指定CUDACXX、CUDA_PATH和/或-DCMAKE_CUDA_COMPILER）

要启用ROCM后端，请添加-DMLIR_ENABLE_ROCM_RUNNER=1并从cmake参数中移除-DLLVM_TARGETS_TO_BUILD="host"。（您可能需要指定-DHIP_CLANG_INCLUDE_PATH和/或ROCM_PATH）

对于启用ISL的polymer，必须将polly添加到LLVM_ENABLE_PROJECTS变量中。

为了加快编译速度，我们建议使用-DLLVM_USE_LINKER=lld。

2. 构建Polygeist：

mkdir build
cd build
cmake -G Ninja .. \
  -DMLIR_DIR=$PWD/../llvm-project/build/lib/cmake/mlir \
  -DCLANG_DIR=$PWD/../llvm-project/build/lib/cmake/clang \
  -DLLVM_TARGETS_TO_BUILD="host" \
  -DLLVM_ENABLE_ASSERTIONS=ON \
  -DCMAKE_BUILD_TYPE=DEBUG
ninja
ninja check-polygeist-opt && ninja check-cgeist

为了加快编译速度，我们建议使用-DPOLYGEIST_USE_LINKER=lld。

GPU后端

要启用CUDA后端，请添加-DPOLYGEIST_ENABLE_CUDA=1

要启用ROCM后端，请添加-DPOLYGEIST_ENABLE_ROCM=1

Polymer

要启用polymer，请添加-DPOLYGEIST_ENABLE_POLYMER=1

可以构建两种polymer配置 - 一种使用Pluto，另一种使用ISL。

Pluto

添加-DPOLYGEIST_POLYMER_ENABLE_PLUTO=1 这将导致cmake调用拉取并构建polymer的依赖项。要指定依赖项的自定义目录，请指定-DPOLYMER_DEP_DIR=<绝对路径>。依赖项将使用tools/polymer/build_polymer_deps.sh进行构建。

要运行polymer pluto测试，请使用ninja check-polymer。

ISL

添加-DPOLYGEIST_POLYMER_ENABLE_ISL=1 这需要一个启用了polly作为子项目的llvm-project构建。

选项2：使用统一的LLVM、MLIR、Clang和Polygeist构建

Polygeist也可以作为外部LLVM项目使用LLVM_EXTERNAL_PROJECTS进行构建。

1. 构建LLVM、MLIR、Clang和Polygeist：

mkdir build
cd build
cmake -G Ninja ../llvm-project/llvm \
  -DLLVM_ENABLE_PROJECTS="clang;mlir" \
  -DLLVM_EXTERNAL_PROJECTS="polygeist" \
  -DLLVM_EXTERNAL_POLYGEIST_SOURCE_DIR=.. \
  -DLLVM_TARGETS_TO_BUILD="host" \
  -DLLVM_ENABLE_ASSERTIONS=ON \
  -DCMAKE_BUILD_TYPE=DEBUG
ninja
ninja check-polygeist-opt && ninja check-cgeist

ninja check-polygeist-opt运行Polygeist/test/polygeist-opt中的测试 ninja check-cgeist运行Polygeist/tools/cgeist/Test中的测试

引用Polygeist

如果您使用Polygeist，请考虑引用以下相关出版物：

@inproceedings{polygeistPACT,
  title = {Polygeist: Raising C to Polyhedral MLIR},
  author = {Moses, William S. and Chelini, Lorenzo and Zhao, Ruizhe and Zinenko, Oleksandr},
  booktitle = {Proceedings of the ACM International Conference on Parallel Architectures and Compilation Techniques},
  numpages = {12},
  location = {Virtual Event},
  series = {PACT '21},
  publisher = {Association for Computing Machinery},
  year = {2021},
  address = {New York, NY, USA},
  keywords = {Polygeist, MLIR, Polyhedral, LLVM, Compiler, C++, Pluto, Polly, OpenScop, Parallel, OpenMP, Affine, Raising, Transformation, Splitting, Automatic-Parallelization, Reduction, Polybench},
}
@inproceedings{10.1145/3572848.3577475,
  author = {Moses, William S. and Ivanov, Ivan R. and Domke, Jens and Endo, Toshio and Doerfert, Johannes and Zinenko, Oleksandr},
  title = {High-Performance GPU-to-CPU Transpilation and Optimization via High-Level Parallel Constructs},
  year = {2023},
  isbn = {9798400700156},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  url = {https://doi.org/10.1145/3572848.3577475},
  doi = {10.1145/3572848.3577475},
  booktitle = {Proceedings of the 28th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming},
  pages = {119–134},
  numpages = {16},
  keywords = {MLIR, polygeist, CUDA, barrier synchronization},
  location = {Montreal, QC, Canada},
  series = {PPoPP '23}
}
@inproceedings{10444828,
  author = {Ivanov, Ivan R. and Zinenko, Oleksandr and Domke, Jens and Endo, Toshio and Moses, William S.},
  booktitle = {2024 IEEE/ACM International Symposium on Code Generation and Optimization (CGO)},
  title = {Retargeting and Respecializing GPU Workloads for Performance Portability},
  year = {2024},
  volume = {},
  issn = {},
  pages = {119-132},
  doi = {10.1109/CGO57630.2024.10444828},
  url = {https://doi.ieeecomputersociety.org/10.1109/CGO57630.2024.10444828},
  publisher = {IEEE Computer Society},
  address = {Los Alamitos, CA, USA},
  month = {mar}
}