IGAnets: Isogeometric analysis networks

IGAnets is a novel approach to combine the concept of deep operator learning with the mathematical framework of isogeometric analysis.

IGAnets requires a C++20 compiler, CMake, and LibTorch (the C++ API of PyTorch).

Supported CMake flags:

• -DIGANET_BUILD_CPUONLY=ON builds IGAnets in CPU mode even if CUDA, ROCm, etc. is found (default OFF).

• -DIGANET_BUILD_DOCS=ON builds the documentation (default OFF). To build the documentation you need Doxygen and Sphinx installed on your system.

• -DIGANET_BUILD_PCH=ON builds IGAnets with precompiled headers (default ON).

• -DIGANET_OPTIONAL="module1[branch];module2[branch];..." builds optional modules (default NONE)

  Optional modules are downloaded into the directory optional. If the current IGAnets checkout is a git repository (i.e. if CMake finds the directory .git) optional modules are also checked out as git repositories. Otherwise, CMake downloads the ZIP archive of the optional module.

  The following optional modules are available:

  • Examples examples[main]
  • Unit tests unittests[main]
  • Performance tests perftests[main]
  • Python bindings python[main]
  • MATLAB bindings matlab[main]

  If [branch] is not given then [main] is assumed by default. There might exist further optional modules that are not visible publicly.

• -DIGANET_WITH_GISMO=ON compiles IGAnets with support for the open-source Geometry plus Simulation Modules library G+Smo enabled (default OFF).

• -DIGANET_WITH_MATPLOT=ON compiles IGAnets with support for the open-source library Matplot-cpp enabled (default OFF). Note that this option can cause compilation errors with GCC.

• -DIGANET_WITH_MPI=ON compiles IGAnets with MPI support enabled (default OFF).

• -DIGANET_WITH_OPENMP=ON compiles IGAnets with OpenMP support enabled (default ON). Note that this option can cause compilation errors with Clang.

1. Install prerequisites (CMake and LibTorch)

   Ubuntu

   apt-get install build-essential cmake unzip wget

   RedHat

   yum install make cmake gcc gcc-c++ unzip wget

   Install LibTorch from PyTorch.org

   Pre-compiled versions of LibTorch are available at PyTorch.org.

   wget https://download.pytorch.org/libtorch/cpu/libtorch-shared-with-deps-2.13.0%2Bcpu.zip -O libtorch.zip
   unzip libtorch.zip -d $HOME/
   rm -f libtorch.zip
   export TORCH_DIR=${HOME}/libtorch/share/cmake/Torch

   Note that there might be a newer LibTorch version available than indicated in the above code snippet.

   Install LibTorch from PyPI

   Alternatively, you can install LibTorch from PyPI using pip and set the TORCH_DIR environment variable:

   python3 -m venv $HOME/.venv/torch-cpu
   source $HOME/.venv/torch-cpu/bin/activate
   pip install torch
   export TORCH_DIR="$(python3 -c \
         'from pathlib import Path; import torch; print(Path(torch.__file__).parent / "share" / "cmake" / "Torch")')"

2. Configure

   cmake .. -DTorch_DIR=${TORCH_DIR}

3. Compile

   make -j 8

   Depending on the number of cores of your CPU you may want to change 8 to a different number.

1. Install prerequisites (CMake and LibTorch)

   brew install cmake pytorch

2. Configure

   cmake .. -DTorch_DIR=/opt/homebrew/Cellar/pytorch/2.13.0/share/cmake/Torch

   Note that the specific version of PyTorch might be different on your system.

3. Compile

   make -j 8

   Depending on the number of cores of your CPU you may want to change 8 to a different number.

1. Install the CUDA-enabled version of LibTorch either through PyTorch.org

   wget https://download.pytorch.org/libtorch/cu126/libtorch-shared-with-deps-2.13.0%2Bcu126.zip -O libtorch.zip
   unzip libtorch.zip -d $HOME/
   rm -f libtorch.zip
   export TORCH_DIR=${HOME}/libtorch/share/cmake/Torch

   or from PyPI

   python3 -m venv $HOME/.venv/torch-cuda
   source $HOME/.venv/torch-cuda/bin/activate
   pip install torch --index-url https://download.pytorch.org/whl/cu126
   export TORCH_DIR="$(python3 -c \
         'from pathlib import Path; import torch; print(Path(torch.__file__).parent / "share" / "cmake" / "Torch")')"

   Note that the version must be compatible with the CUDA version installed on your system.

2. Configure and compile

   All further steps are the same as described above (Linux)

1. Install the ROCm-enabled version of LibTorch either through PyTorch.org

   wget https://download.pytorch.org/libtorch/rocm7.2/libtorch-shared-with-deps-2.13.0%2Brocm7.2.zip -O libtorch.zip
   unzip libtorch.zip -d $HOME/
   rm -f libtorch.zip
   export TORCH_DIR=${HOME}/libtorch/share/cmake/Torch

   or from PyPI

   python3 -m venv $HOME/.venv/torch-rocm
   source $HOME/.venv/torch-rocm/bin/activate
   pip install torch --index-url https://download.pytorch.org/whl/rocm7.2
   export TORCH_DIR="$(python3 -c \
         'from pathlib import Path; import torch; print(Path(torch.__file__).parent / "share" / "cmake" / "Torch")')"

   Note that the version must be compatible with the ROCm version installed on your system.

2. Configure and compile

   All further steps are the same as described above (Linux)

1. Install the Intel GPU drivers and PyTorch version as described here. If you do not own an Intel GPU, you can create a free account at the Intel Tiber AI Cloud, which provides free access to Intel data-center GPUs for testing purposes.

2. Install the XPU-enabled version of PyTorch in a virtual python environment

   python3 -m venv $HOME/.venv/torch-xpu
   source $HOME/.venv/torch-xpu/bin/activate
   pip install torch --index-url https://download.pytorch.org/whl/xpu
   export TORCH_DIR="$(python3 -c \
         'from pathlib import Path; import torch; print(Path(torch.__file__).parent / "share" / "cmake" / "Torch")')"

3. Configure

   cmake .. -DTorch_DIR=${TORCH_DIR}

   Note that on the latest Intel Tiber AI Cloud installation, ZLib is not found by default. This can be corrected by calling CMake with the additional parameters

   -DZLIB_LIBRARY=/usr/lib/x86_64-linux-gnu/libz.so.1 -DZLIB_INCLUDE_DIR=/usr/include

1. Install the Intel Extensions for PyTorch as described here.

2. Add the CMake option -DIPEX_DIR=<path/to/IPEX/installation>