A non-linear finite element code. This project is still under development and is not considered stable.
Before using the package, the code needs to be compiled and installed on the local system. There are two ways of accomplishing this task; using a docker image in a secure environment with the dependencies already handled, or compiling this on the host operating system.
Go to the top level directory in the git repository and enter
sudo docker build -t neon .
This will pull down the base image and then clone the git repository inside the docker container. The original hosting can be found at https://hub.docker.com/r/dbeurle/neon/
Once this is completed, enter the following commands to enter the docker container
$ sudo docker run -i -t neon /bin/bash
The instructions in the section below can now be followed with the copy of the repository inside the docker container without worrying about dependencies.
The external dependencies are:
- Boost filesystem (to be replaced with
std::filesystem) - Pastix and MUMPS for direct linear solvers
- VTK for writing out simulation data to a ParaView compatible format
- An OpenMP enabled C++17 compiler
- Intel Thread Building Blocks TBB
Other dependencies are pulled in during build time with CMake and include
- Eigen for linear algebra
- range-v3 for range support
- Termcolor for colour terminal support
- Catch for unit testing
For best performance build with native optimisations on for the machine you are using. This will automatically trigger Eigen to use wider SIMD instruction when available on native architecture at the expense of portability.
The build instructions for debug are
$ mkdir build && cd build/$ cmake ..$ make all
and for release mode with optimisations
$ mkdir build && cd build/$ cmake -DCMAKE_BUILD_TYPE=Release ..$ make all -j<# cpus>$ sudo make install
If the code is only used on a particular machine, then machine specification optimisations can be invoked by specifying the CMake symbol
-DENABLE_NATIVE=1
which enables the compiler to generate the best machine code for your platform.
If you have an NVIDIA graphics card, then you can use the CUDA enabled iterative solvers by specifying the CMake symbol
-DENABLE_CUDA=1
CUDA doesn't follow the latest GCC so a compatibility package is usually provided. CMake can be told to use a specific compiler for the host compilation through
-DCMAKE_CUDA_HOST_COMPILER=<compiler>
For checking the successful compilation of the program, invoke the test suite by executing
$ ctest
in the build directory.
Install dependencies through the package manager:
$ sudo apt install cmake git mercurial zlib1g-dev libcurl4-openssl-dev libvtk6-dev libtbb-dev libboost-filesystem-dev libmumps-seq-dev libopenblas-dev libarpack2-dev libscotch-dev hwloc libhwloc-dev gfortran
Then clone the repository and add
$ git clone https://github.com/dbeurle/neon.git
and enter the repository
$ cd neon/docker && sh install_pastix.sh
After this compiles and fails to install, enter the commands to install and link the libraries
$ cd pastix_5.2.3/build && sudo make install && sudo ln -s /usr/local/lib/libpastix.so /usr/lib/libpastix.so
Provide the sudo password when prompted.
Now we are going to create the build directory and compile. Navigate back to the neon directory and create and enter the build directory
$ mkdir build && cd build
We can create the compilation environment and setup optimisations using CMake with
$ cmake -DCMAKE_BUILD_TYPE=Release ..
Finally compiling with
$ make all -jN
where N is the number of parallel build jobs you want to run.
You can then install using
$ sudo make install
and run the test suite with
$ ctest -jN
to ensure the program is behaving as expected.
See the LICENSE.md file for the project license and the licenses of the included dependencies.
Many thanks to the contributors of ideas, code and theoretical discussions:
- Shadi Alameddin (code reviews, verification, bug reports, theoretical and code contributions, damage model)
- Shannon Beurle (documentation, quality, beam section properties and tests)
- Patricio Rodrigeuz (pyramid shape functions and quadrature)
If you are missing please open an issue and I'll happily add you to the list.