An open-source platform for computational mechanics. It uses Kokkos and will run in parallel (MPI) and on the GPU.
aperi-mech depends on:
- Trilinos packages including:
kokkos,kokkos-kernels,exodus,STK, and more. - Python packages include:
NumPy,SciPy,matplotlib,jupyter, and mores. - And other:
googletest,yaml,eigen
aperi-mech uses Git Large File Storage (LFS) for managing large files. Ensure you have git-lfs installed and configured before cloning the repository.
Follow the instructions at https://git-lfs.github.com/. In short:
# Install git-lfs
sudo apt-get install git-lfs
# or
brew install git-lfs
# Initialize git-lfs in your repository
git lfs installWhen cloning the repository, ensure that LFS files are pulled correctly:
git clone https://github.com/aperijake/aperi-mech.git
cd aperi-mech
git lfs pullDockerfiles are provided for the easiest way to get up and running. Otherwise, the recommended way to build is using spack and conda or pip.
See https://www.docker.com/get-started/ to install Docker on your system. After that, you can build an MPI CPU image:
docker build -f Dockerfile -t aperi-mech:latest .and start the container interactively using:
docker run -it aperi-mech:latestFor a GPU build, see the other Dockerfile* files in this directory. Be sure to change the CUDA_ARCH to match your GPU. The Dockerfile can also be used to see how the project is built.
Follow the instructions at https://spack.readthedocs.io/en/latest/getting_started.html. In short, install the spack prerequisites, clone spack, and source the script for shell support.
Next find the system tools:
# Add system compilers. e.g. gcc
spack compiler find
# Use external tools. e.g. cmake
spack external find# Create and activate a spack environment for the project
spack env create aperi-mech
spacktivate aperi-mechFor a CPU build with MPI:
# If needed, specify a specific compiler. For example, add `%[email protected]` at the end of the `spack add` commands
# Requires a Trilinos commit from 04/19/2024, so >[email protected]
spack add compadre
spack add kokkos ~cuda ~shared cxxstd=17
spack add kokkos-kernels ~cuda ~shared
spack add trilinos@develop ~amesos ~amesos2 ~anasazi ~aztec ~belos ~cuda ~epetra ~epetraext ~ifpack ~ifpack2 ~ml ~muelu ~sacado ~shared +exodus +gtest +hdf5 +stk +zoltan +zoltan2 cxxstd=17
spack add googletest
spack add lcov
spack add yaml-cpp
spack add eigen
# Install Packages
spack installOr, for a GPU build:
# Replace cuda_arch according to your GPU.
spack add compadre
spack add kokkos-kernels +cuda ~shared cuda_arch=75
spack add kokkos +cuda +cuda_lambda +cuda_relocatable_device_code ~cuda_uvm ~shared +wrapper cuda_arch=75 cxxstd=17
spack add trilinos@develop ~amesos ~amesos2 ~anasazi ~aztec ~belos ~cuda ~epetra ~epetraext ~ifpack ~ifpack2 ~ml ~muelu ~sacado ~shared +cuda +cuda_rdc +exodus +gtest +hdf5 +stk +zoltan +zoltan2 cuda_arch=75 cxxstd=17
spack add googletest
spack add lcov
spack add yaml-cpp
spack add eigen
# Install Packages
spack installFollow instructions here https://docs.conda.io/projects/conda/en/stable/user-guide/install/index.html. In short:
# For example, replace with the appropriate system file
wget https://repo.anaconda.com/archive/Anaconda3-2024.06-1-Linux-x86_64.sh # download
bash Anaconda3-2024.06-1-Linux-x86_64.sh # install
# open fresh terminal
conda init bash # initialize, may have to path to conda if it is not found
conda update conda
conda config --add channels conda-forge # add path to conda-forgeCreate an environment
conda create -n aperi-mech # change name as desired
conda activate aperi-mechInstall required packages
conda install pytest testbook jupyter jupyterlab numpy scipy matplotlib ipykernelAdd the environment as a Jupyter kernel so it can be used to run the notebooks
python -m ipykernel install --user --name aperi-mech --display-name "Python (aperi-mech)"From the project root directory with the spack environment activated, configure with cmake
./do_configurenote the options in do_configure script to enable GPU support (--gpu) and switch the build type (e.g. --build-type Debug)
cd build/Release # for CPU and release
# cd build/Release_gpu # for GPU and release
# cd build/Debug # for CPU and debug
# cd build/Debug_gpu # for GPU and debug
# Build the project
make -j 4 # change 4 to the number of processors desiredSuccessfully installed on an Azure NC4as T4 v3, Ubuntu 20.04, system using apt-get to install prerequisites:
The commands for a CPU build with OpenMP are below. There was some trouble building Compadre tests, so the ~tests flag was important.
# Add packages
spack add kokkos%[email protected] ~cuda ~shared +openmp cxxstd=17
spack add kokkos-kernels%[email protected] ~cuda ~shared +openmp
spack add compadre@master%[email protected] ~tests
spack add trilinos@master%[email protected] +boost ~cuda +exodus +gtest +hdf5 ~muelu +openmp ~sacado ~shared +stk +zoltan +zoltan2 cxxstd=17 # This can probably be trimmed more
spack add googletest%[email protected]
spack add yaml-cpp%[email protected]
spack add eigen%[email protected]
spack add lcov%[email protected]
# Install
spack install --freshSuccessfully installed on an Ubuntu 23.10 system using apt-get to install prerequisites:
Successfully installed on a M1 Mac with the following:
[email protected]from Homebrew[email protected]from Xcode[email protected]_1from Homebrew
For code coverage, be sure gcov is using from the same version of gcc. I had to:
ln -s /opt/homebrew/bin/gcov-13 /opt/homebrew/bin/gcovAlso, some homebrew installed tools were not in the path by default and had to be added or some package would fail to install.
Spack struggled to install lcov due to issues installing perl so used the homebrew lcov instead
# Add packages
spack add compadre@master%[email protected]
spack add kokkos-kernels%[email protected] -cuda -shared +threads
spack add kokkos%[email protected] -cuda -shared +threads cxxstd=17
spack add trilinos@master%[email protected] -cuda +exodus +gtest +hdf5 -shared +stk +zoltan +zoltan2 cxxstd=17
spack add googletest%[email protected]
spack add yaml-cpp%[email protected]
spack add eigen%[email protected]
# Install Packages
spack install