Skip to content

Latest commit

 

History

History
253 lines (206 loc) · 17.6 KB

README.md

File metadata and controls

253 lines (206 loc) · 17.6 KB
Raw

ExaCA

An exascale-capable cellular automaton for nucleation and grain growth

ExaCA is a cellular automata (CA) code for grain growth under additive manufacturing conditions, created by ExaAM within the Exascale Computing Project.

Build

ExaCA uses Kokkos and MPI for parallelism and JSON for input files.

Dependencies

Dependency Version Required Details
CMake 3.11+ Yes Build system
Kokkos 4.0+ Yes Portable on-node parallelism
MPI GPU Aware if CUDA/HIP Enabled Yes Multi-node parallelism
json 3.10+ Yes Input parsing
GoogleTest 1.10+ No Unit test framework
CUDA 9+ No Programming model for NVIDIA GPUs
HIP 3.5+ No Programming model for AMD GPUs
Finch main No Heat transport model for coupled simulation

CMake must be available to build ExaCA and Kokkos. The underlying parallel programming models and MPI are available on most systems and can generally be found automatically by CMake. Note these dependencies must all be installed first, if not available. Kokkos is also available on many systems; if not, obtain the desired version:

git clone https://github.com/kokkos/kokkos.git --branch 4.0.00

Obtaining ExaCA

ExaCA is available on GitHub, by default starting from the current master branch:

git clone https://github.com/LLNL/ExaCA.git

Backend options

Note that ExaCA runs with the default enabled Kokkos backend (see https://github.com/kokkos/kokkos/wiki/Initialization).

ExaCA has been tested with the Serial, OpenMP, Threads, CUDA, and HIP backends.

Building Kokkos

Host build

If Kokkos is not already installed on your system, configure, build, and install Kokkos:

# Change this path to desired Kokkos installation location
export KOKKOS_INSTALL_DIR=`pwd`/install/kokkos
# Change this path to Kokkos source
cd ./kokkos
# Configure Kokkos
cmake \
  -B build \
  -D CMAKE_BUILD_TYPE="Release" \
  -D CMAKE_INSTALL_PREFIX=$KOKKOS_INSTALL_DIR \
  -D Kokkos_ENABLE_OPENMP=ON
# Build Kokkos
cmake --build build
# Install Kokkos
cmake --install build
cd ../

Note that there are other host backends available. Kokkos architecture flags can also be set to improve performance and must match the hardware you run on (e.g. -DKokkos_ARCH_POWER9=ON); see Kokkos architecture flag details.

CUDA build

Similar to the CPU build above, Kokkos can instead be configured and built for NVIDIA GPUs:

# Change this path to desired Kokkos installation location
export KOKKOS_INSTALL_DIR=`pwd`/install/kokkos
# Change this path to Kokkos source
cd ./kokkos
# Check the GPU architecture flag matches the hardware
# Configure Kokkos
cmake \
  -B build \
  -D CMAKE_BUILD_TYPE="Release" \
  -D CMAKE_INSTALL_PREFIX=$KOKKOS_INSTALL_DIR \
  -D Kokkos_ENABLE_CUDA=ON \
  -D Kokkos_ENABLE_CUDA_LAMBDA=ON \
  -D Kokkos_ARCH_VOLTA70=ON
# Build Kokkos
cmake --build build
# Install Kokkos
cmake --install build
cd ../

Note the two flags needed for the Kokkos::Cuda backend. The Kokkos architecture flag must match the hardware you run on. Kokkos will automatically redirect the (default) host compiler to nvcc in the example above. By default, the host will use Kokkos::Serial; other parallel host backends can also be used, e.g. by adding -D Kokkos_ENABLE_OPENMP as was done above.

HIP Build

To build Kokkos for HIP the hipcc compiler must be explicitly passed, along with architecture and backend flags analogous to the previous examples:

cd ./kokkos
# Configure Kokkos
cmake \
  -B build \
  -D CMAKE_BUILD_TYPE="Release" \
  -D CMAKE_CXX_COMPILER=hipcc \
  -D CMAKE_INSTALL_PREFIX=install \
  -D Kokkos_ENABLE_HIP=ON \
  -D Kokkos_ARCH_VEGA908=ON
# Build Kokkos
cmake --build build
# Install Kokkos
cmake --install build
cd ../

Build ExaCA

Once Kokkos and MPI are installed, ExaCA can be built:

# Change this path to desired ExaCA installation location
export EXACA_INSTALL_DIR=`pwd`/install/exaca
# Change this path to Kokkos installation location
export KOKKOS_INSTALL_DIR=`pwd`/install/kokkos

# Change this path to ExaCA source
# Configure ExaCA
cd ./ExaCA
cmake \
  -B build \
  -D CMAKE_BUILD_TYPE="Release" \
  -D CMAKE_PREFIX_PATH=$KOKKOS_INSTALL_DIR \
  -D CMAKE_INSTALL_PREFIX=$EXACA_INSTALL_DIR
# Build ExaCA
cmake --build build
# Install ExaCA
cmake --install build
cd ../

Kokkos will forward the compilation flags and details to ExaCA automatically.

Building with external JSON

By default, ExaCA will download the JSON library dependency used for input files. This automatic download does not work on all systems; a separate build of this library can be done instead. As with the dependencies described above, first obtain the source:

git clone https://github.com/nlohmann/json

And then build the json library (header only):

# Change this path to desired JSON installation location
export JSON_INSTALL_DIR=`pwd`/install/json

cd json
# Configure json
cmake \
  -B build \
  -D CMAKE_BUILD_TYPE="Release" \
  -D CMAKE_INSTALL_PREFIX=$JSON_INSTALL_DIR \
  -D JSON_BuildTests=OFF
# Build json
cmake --build build
# Install json
cmake --install build
cd ../

Then add this install path to the ExaCA configuration (example above) together with the path to Kokkos -D CMAKE_PREFIX_PATH="$KOKKOS_INSTALL_DIR;$JSON_INSTALL_DIR" and build ExaCA. Note that quotes are necessary for multiple paths.

Building with Finch

ExaCA can be compiled with Finch, a finite difference-based heat transport solver, for coupled heat transport and solidification simulation without the need to read time-temperature history data from file(s). The Finch source code and build instructions are available at https://github.com/ORNL-MDF/Finch. To compile ExaCA with Finch, include the path to the Finch install in the CMAKE_INSTALL_PREFIX. To require that ExaCA is compiled with Finch, add ExaCA_REQUIRE_FINCH=ON.

Testing ExaCA

Unit tests can be run if the ExaCA_ENABLE_TESTING CMake option is enabled in the build and if the GoogleTest framework is available on the system or built locally with the install path passed to ExaCA (see the previous section describing the JSON build and pointing ExaCA to the installation).

After building, tests can be run with cmake --build build --target test from the source directory (presuming build/ is the relative location of the build folder). Tests are automatically generated for all enabled Kokkos backend.

Running ExaCA

ExaCA runs using an input file, passed on the command line. Example problems are provided in the examples/ directory - a separate examples/README.md file goes into more detail on the problem types, the optional and required arguments needed for each problem type, and additional files used by ExaCA. The example input files present in this repository are:

  • Inp_DirSolidification.json: simulates grain growth from a surface with a fixed thermal gradient and cooling rate
  • Inp_SmallDirSolidification.json: a smaller and simpler version of the previous
  • Inp_SpotMelt.json: simulates overlapping spot melts with fixed a fixed thermal gradient and cooling rate
  • Inp_SmallSpotMelt.json: a smaller and simpler version of the previous

Example problems only possible with external data:

  • Inp_SingleLine.json: simulates melting and solidification of a single line of melt pool data
  • Inp_SingleLineTranslate.json: using the same single line data as in Inp_SingleLine.json, create 3 overlapping line melt pools using specified bounds in the Y direction
  • Inp_TwoLineTwoLayer.json: simulates two layers consisting of segments of two overlapping melt pools

Run by calling the created executable with an ExaCA input file:

mpiexec -n 1 ./build/install/bin/ExaCA examples/Inp_DirSolidification.json

Alternatively, the Finch-ExaCA executable can be used for coupled Finch-ExaCA simulations. Compilation with ExaCA_ENABLE_FINCH=ON is required for these examples with Finch in-memory coupling:

  • Inp_SmallFinch.json: simulates melting and solidification of a small melt pool segment at a coarse resolution
  • Inp_Finch.json: simulates melting and solidification of a small melt pool segment at a fine resolution, repeated for 3 layers
  • Inp_FinchTranslate.json: simulates melting and solidification of the small melt pool segment at the fine resolution translated in space to form 3 overlapping segments

For coupled Finch-ExaCA runs, caution should be taken such that the cell size given in the CA input file matches that given in the Finch input file. Additionally, scan_path_file in the Finch input file (for Inp_Finch.json and Inp_FinchTranslate.json, this is examples/single_line/inputs_small_refined.json in the Finch repository and for Inp_SmallFinch.json, this is examples/single_line/inputs_small.json in the Finch repository) should be modified to represent a global path name. Run by calling the created executable with a Finch input file and an ExaCA input file on the command line, with the Finch input file listed first:

mpiexec -n 1 ./build/install/bin/Finch-ExaCA $PATH_TO_FINCH/examples/single_line/inputs_small.json examples/Inp_SmallFinch.json

Automated input file generation using Tasmanian (https://tasmanian.ornl.gov/)

Within the utilities directory, an example python script for the generation of an ensemble of input files is available. By running the example script TasmanianTest.py, 69 ExaCA input files are generated with a range of heterogeneous nucleation density, mean nucleation undercooling, and mean substrate grain size values, based on the ranges in python code (N0Min-N0Max, dTNMin-dTNMax, and S0Min-S0Max), respectively. Running the python script from the ExaCA source directory, via the command

python utilities/TasmanianTest.py PathToTemperatureFile1 PathToTemperatureFile2 ...

the script will generate an ensemble of input files in the examples directory, for a series of simulations that will use the thermal history or histories described in PathToTemperatureFile1(s) being repeated for a certain number of layers (56 in this example). If a simulation repeating multiple thermal histories is desired (for example, and even layer and an odd layer scan pattern), both paths to/file names of the thermal history data should be given on the command line. Running this code will generate N = 1 to 69 input files named examples/Inp_TasmanianTest_[N].json. Other CA inputs, such as the time step or cell size, must be adjusted manually inside of the python script. Separate instances of ExaCA can be run with each ensemble member to probe microstructure dependency on nucleation and substrate.

Output and post-processing analysis

As detailed further in examples/README.md, ExaCA can print output fields from the simulation at either a specified increment during a simulation (or during a given layer of a multilayer problem), or at the end of the simulation (or after specified layers of a multilayer problem). A list of output options is given below, along with descriptions and whether each field is stored for all layers of a multilayer problem or only the current simulated layer.

Output field Stored for Details
GrainID All layers A unique identifier for each grain in the ExaCA simulation. Positive numbers represent grains from either the initial substrate/baseplate or from a powder layer, while negative numbers represent grains formed via a nucleation event. GrainIDs of 0 are used to identify cells from a powder layer that do not have any associated material (i.e., voids or gas in the powder layer). Each GrainID is linked to a grain orientation through the substrate input file - GrainID = 1 or -1 is associated with the first orientation in the file, GrainID = 2 or -2 with the second orientatation, etc, with orientations reused when the number of grains exceeds the number of orientations in the file (for example, examples/Substrate/`GrainOrientationVectors.csv has 10,000 grain orientations, so a GrainID of 10,001 will share its orientation with the GrainID of 1).
LayerID All layers The final (or most recent) layer in which a given cell undergoes/underwent melting and solidification. Layer numbers start at 0, with -1 used for cells that have not undergone melting/solidification.
GrainMisorientation All layers The orientation (in degrees) of the grain's nearest <100> direction to the Z axis (the thermal gradient direction for directional solidification problems, the build/layer offset direction for other problems). Epitaxial grains (from the initial grain structure or powder layer) are assigned (rounded) integer values between 0 and 55, while nucleated grains (not present in the initial grain structure) are assigned values between 100 and 155 (the offset of 100 is simply used to ensure the two types of grains are differentiated, but a nucleated grain with a value of 135 actually has <100> misoriented with Z by 35 degrees). All cells belonging to the same grain (same GrainID value) will have the same GrainMisorientation, as no intra-granular misorientations are currently simulated. Also of note is that for intermediate output files where liquid cells are present, these cells are assigned a value of -1, and that unmelted material (or non-existent material in a powder layer) is assigned a value of 200.
UndercoolingCurrent All layers The current undercooling of a cell (if liquid or active), or the undercooling when a cell was most recently active (if solid). Superheated liquid cells or cells that have not undergone melting and solidification will have a value of 0.
UndercoolingSolidificationStart All layers For cells that are currently undergoing solidification or have finished solidification, the undercooling when solidification started. Will have a value of 0 for all other cells
MeltPoolEdge All layers Value of 1 for cells that were at the edge of the time-temperature history data and converted to active cells, value of 0 for all other cells including cells that became active via a cell capture or nucleation event
MeltTimeStep Current layer The time step at which a cell will go above the liquidus for the final time (either overall, or for a multilayer problem, in the current layer). Will have a value of -1 for cell that will not go above the liquidus.
CritTimeStep Current layer The time step at which a cell will go below the liquidus for the final time (either overall, or for a multilayer problem, in the current layer). Will have a value of -1 for cell that will not go above the liquidus.
UndercoolingChange Current layer The rate (in K per time step) at which a cell will cool below the liquidus when it undergoes solidification for the final time (either overall, or for a multilayer problem, in the current layer). Will have a value of -1 for cell that will not go above the liquidus.
CellType Current layer The state of a given ExaCA cell, mapped from integers to values according to the types given in src/CAtypes.hpp
DiagonalLength Current layer The size of the octahedral envelope used to track the solidification front (will be 0 if a cell has not yet/will not undergo solidification).
SolidificationEventCounter Current layer The number of times a cell has undergone solidification (either overall, or for a multilayer problem, in the current layer).
NumberOfSolidificationEvents Current layer The number of times a cell will undergo solidification (either overall, or for a multilayer problem, in the current layer).

Analysis of ExaCA vtk data from the final state of a problem can be performed if the vtk data contains, at a minimum, the LayerID and GrainID fields. For example, using output for the test problem Inp_DirSolidification.json (output files TestProblemDirS.vtk and TestProblemDirS.json), analysis can be performed using the ExaCA-GrainAnalysis executable (installed in the same location as ExaCA), with two command line arguments: the first being the path to/name of the analysis input file, and the second being the path to and filename (excluding extensions) of the .vtk and .json files associated with the data set of interest.

./build/install/bin/ExaCA-GrainAnalysis analysis/examples/AnalyzeDirS.json TestProblemDirS

Within the analysis/examples directory, there are example analysis input files. Note that the microstructure data files TestProblemDirS.vtk and TestProblemDirS.json must both be in the location given on the command line.

The analysis executable, in addition to outputting grain statistics, can also output files that can be further post-processing in Matlab using the MTEX toolbox to generate pole figures, inverse pole figures, and inverse pole figure-colored cross-sections. More details on this are provided in analysis/README.md

Citing ExaCA

If you use ExaCA in your work, please cite the following paper. In addition, cite the current release or version used from Zenodo.

Publications using ExaCA

In addition to the primary ExaCA citation, this list of articles use ExaCA in their work.

Contributing

We encourage you to contribute to ExaCA. Please check the contribution guidelines.

License

ExaCA is distributed under an MIT license.

Release

LLNL-CODE-821827