DARTS build instructions

Windows (multithread)

  1. Generate git keys pair and upload the public key to GitLab

  2. Clone the DARTS project from GitLab

  3. Install Python 3.6 or higher. Set the PYTHONPATH environment variable. You should restart VS to take them updated in it. Add path to Python to your environment variable PATH (for example C:\WPy64-39100\python-3.9.10.amd64)

  4. Open the Visual Studio 2022 solution darts-engines.sln in the directory darts-engines

  5. Select the configuration you want (Release_MT for multithreaded Release) and build the solution.

  6. To install DARTS modules to your Python distribution: run WinPython-64bit-3.6.0.1Qt5\WinPython Command Prompt.exe then run darts\darts-package\build_install_darts.bat in this console.

Windows (multithread+GPU)

  1. Install CUDA Toolkit and get AMGX. You can get AMGX sources as a submodule of darts-linear-solvers, then you will found it in darts-linear-solvers\lib\AMGX If you want to build AMGX yourself: run cmake to configure and create VS projects. Example: cmake -DCUDA_ARCH="70" -DCMAKE_BUILD_TYPE=Release -A x64  -B "./AMGX/build"
    If needed add -DCMAKE_CUDA_COMPILER="C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.2/bin/nvcc.exe" Ready to use amgxsh.dll can be downloaded via the link Copy amgxsh.dll and all Cuda DLL files from C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\bin to the python package directory, for example WPy64-39100\python-3.9.10.amd64\Lib\site-packages\darts

  2. Do the items 1-4 from the Windows (multithread) instruction

  3. Use _gpu.sln and Release_MT_gpu configuration for linear_solvers and darts-engines projects

  4. Do the items 6-10 from Windows (multithread) instruction

Linux (multithread)

  1. Install python3, python3-dev, and pip

  2. Install numpy (pip install numpy)

  3. run build_darts.sh

Linux (multithread+GPU)

  1. Download and install NVIDIA CUDA Toolkit
    Errors may fixed by manually installing liburcu6
    wget http://ftp.de.debian.org/debian/pool/main/libu/liburcu/liburcu6_0.12.2-1_amd64.deb
    sudo dpkg -i liburcu6_0.12.2-1_amd64.deb

  2. Download and build AMGX
    git clone https://github.com/NVIDIA/AMGX.git
    cmake -DCMAKE_BUILD_TYPE=Release -B “./build”
    cd build && make -j 4
    The ‘4’ is number of threads used for parallel compilation. In case you use gcc 11.2 you probably will get a compilation error in std_function.h, use this solution

  3. Do the items 1-4 from the Linux (multithread) instruction with make gpu (multithread with GPU support) instead of make mt (multithread)