ECMech
8888888888 .d8888b. 888b d888 888
888 d88P Y88b 8888b d8888 888
888 888 888 88888b.d88888 888
8888888 888 888Y88888P888 .d88b. .d8888b 88888b.
888 888 888 Y888P 888 d8P Y8b d88P" 888 "88b
888 888 888 888 Y8P 888 88888888 888 888 888
888 Y88b d88P 888 " 888 Y8b. Y88b. 888 888
8888888888 "Y8888P" 888 888 "Y8888 "Y8888P 888 888
ECMech is a GPU-friendly library of constitutive models. The models are based on standard continuum mechanics concepts. Crystal-mechanics-based and porosity-mechanics-based models are a principal focus. Models are meant for standard crystalline metallic materials deforming under quasi-static conditions.
Constitutive model response is a main ingredient in the simulation of deformation of material. In the solution of the governing equations, the constitutive model sub-problem is typically solved at every discretization point in the problem. In a finite element setting for solving the balance of momentum (stress equilibrium in the quasi-static case), these would be quadrature points in the finite element integrals. The constitutive model has two main jobs: it provides the stress tensor that goes into the balance of linear momentum, and it updates the evolving state of the material. This state can be tracked by variables for the stress, dislocation density, orientation of crystal lattices, grain size, and so forth.
Implementations here are meant for quasi-static applications, and include material tangent stiffness computations for incorporation into finite element stiffness matrices for quasi-static applications.
A common feature of advanced constitutive models is the need to solve coupled systems of non-linear equations. These solutions typically occur at each time step at every integration point in the host code, but with computational work at the various integration points being independent and well-suited to parallel computation. These multi-dimensional systems of equations arise in a variety of models including those for crystal-mechanics, porous-plasticity, and phase kinetics. The systems of equations are of comparatively small size; but they are often numerically stiff, requiring robust iterative solution schemes to produce numerically stable results. ECMech makes use of the open-source library snls to solve these systems of equations.
Examples of work that has made use of similar models and algorithms:
- Journal publication on application to an alpha-beta titanium alloy
- Journal publication for comparison to diffraction data
A target platform for ExaCMech is the CORAL platform which for a typical node on a CORAL machine has between 40 - 42 IBM Power9 processors and between 4 - 6 Nvidia V100 GPUs. Additional information on CORAL can be found at: https://hpc.llnl.gov/training/tutorials/using-lcs-sierra-system#CORAL .
The build system is cmake-based.
Dependencies:
- blt -- required
- https://github.com/LLNL/blt
- use
git submodulecommands, or specify with-DBLT_SOURCE_DIR=...
- snls -- required
- https://github.com/LLNL/SNLS
- use
git submodulecommands, or specify with-DSNLS_DIR=...
- raja -- required
- https://github.com/LLNL/RAJA
- A RAJA for the miniapp on a CORAL architecture was built with the following commands
cmake ../ -DCMAKE_INSTALL_PREFIX=../install_dir/ -DENABLE_OPENMP=ON -DENABLE_CUDA=ON -DRAJA_TIMER=chrono -DCUDA_ARCH=sm_70 - specify with
-DRAJA_DIR=.../share/raja/cmake
Run cmake with -DENABLE_TESTS=ON and do make test
The develop branch is the main development branch. Changes to develop are by pull request.
A miniapp is provided which provides a guide for how the library might be used in a real application. It provides several compute kernels that contain CPU, OpenMP, and CUDA/HIP code through the use of RAJA. A bash script is provided for a CORAL machine that runs a few different examples using the CPU, OpenMP, and CUD/HIP for realistic size data.
If all of the features are used it can be built with the following commands when used on a CORAL architecture: cmake ../ -DCMAKE_INSTALL_PREFIX=../install_dir/ -DRAJA_DIR={$RAJA_PATH}/raja/install_dir/share/raja/cmake/ -DENABLE_OPENMP=ON -DENABLE_CUDA=ON -DCUDA_ARCH=sm_70 -DENABLE_TESTS=ON -DENABLE_MINIAPPS=ON -DCMAKE_BUILD_TYPE=Release
An initial set of pythons bindings are now available to users. These bindings allow one to run material point simulation like calculations from within python through the pyECMech class. One can look at pyecmech/example.py as an example for how to use these bindings. In order to enable these bindings, one has to run cmake with -DENABLE_PYTHON=ON.
If one is more interested in development work from within python they can add the additional cmake define -DENABLE_PYTHON_DEV=ON. This cmake define introduces a new python class called pyECMechDev which allows developers access to the internals of getResponseECM. The intended use of this class is currently to allow developers to play around with different solvers for the nonlinear set of equations being solved within ExaCMech. One can look at pyecmech/example-dev.py as an example of how to use these bindings to run material point simulations.
The principal developer of ECMech is Nathan Barton, [email protected]. Robert Carson has also made contributions.
ECMech may be cited using the following bibtex entry:
@Misc{ecmech,
title = {{ECMech}},
author = {Barton, Nathan R. and Carson, Robert A. and Wopschall, Steven R and USDOE National Nuclear Security Administration},
abstractNote = {{ECMech} is a {GPU}-friendly library of constitutive models. Crystal-mechanics-based and porosity mechanics-based models are a principal focus.},
url = {https://github.com/LLNL/ExaCMech},
doi = {10.11578/dc.20190809.2},
year = {2018},
month = {12},
annote = {
https://www.osti.gov//servlets/purl/1550790
https://www.osti.gov/biblio/1550790-llnl-exacmech
}
}
License is under the BSD-3-Clause license. See LICENSE file for details. And see also the NOTICE file.
SPDX-License-Identifier: BSD-3-Clause
LLNL-CODE-784997