feature-extractor

feature-extractor is a LLVM analysis pass to extract static program features for use in machine learning projects. The extracted features are those present in the following article:

Practical Aggregation of Semantical Program Properties for Machine Learning Based Optimization

Table of Contents

• Requirements
  • Platform Requirements
  • Build Requirements
  • Runtime Requirements
• Building from Source
• Contributors
• License

Requirements

Platform Requirements

This pass has only been tested on Linux, although it is expected to run on MacOS without any modifications; Other platforms may require modifications to the source code (contributions are welcome).

Build Requirements

In order to build feature-extractor from source you will need:

• LLVM 9 or higher
• CMake 3.4.3 or higher
• CMake compatible build system like Make or Ninja
• Modern C++14 compiler (Clang is recommended)

Runtime Requirements

In order to run this pass, you will need a compiler that either emits LLVM IR to be used with llvm-opt or can use LLVM out-of-tree passes natively, such as Clang or Rust.

Building from Source

Example using Clang as compiler and Ninja as build system:

export CC=clang
export CXX=clang++
export CMAKE_GENERATOR=Ninja
export CMAKE_BUILD_TYPE=Release
git clone https://github.com/andrefz/feature-extractor.git
cd feature-extractor
mkdir build
cd build
cmake .. && cmake --build .

Contributors

Those are the people who contributed to this project, in alphabetical order:

• @andrefz - André F. Zanella
• @leonardohn - Leonardo H. Neumann

License

This project is licensed under the MIT License.