This repository contains code for a simple CUDA accelerated material point method simulation.

Here is a demo of the simulation which was rendered offline using Blender. (Click to view)

The easiest way to run the code is using the provided Docker configuration. Build the image by running docker build -t mpm . from the project directory. Once built, the image can be run with ./docker_run.sh. Any arguments passed in, will be passed on to the simulation program. Check include/options.h for a full list of arguments.

Here are some example simulations to try:

./docker_run.sh --scene scenes/rubber_duck.toml -N 16 --particle-count 10000
./docker_run.sh --scene scenes/liquid_bunny.toml -N 32 --particle-count 1000000
./docker_run.sh --scene scenes/snowman.toml -N 64 --particle-count 500000

You should see the real-time debug view pop up. It looks 2D, but the simulation is actually run in 3D.

To export meshes add the --save-dir out/ as an option.

./docker_run.sh --save-dir out/

Reconstructed meshes will be written to the directory out/meshes.

Our code was developed on Linux (Ubuntu, specifically). It should work anywhere, of course, but has never been tested on Windows or Mac. The below instructions will therefore be specific to Ubuntu. Please feel free to apply them to the system of your choice analogously.

We use cmake to build stuff.

The following dependencies are required:

• OpenGL
• GLFW 3
• OpenMP (Kind-of required. If not available, set the cmake option USE_OPENMP to OFF)
• Boost
• Eigen3

And these are optional:

• Partio (just for exporting particles)
• GTest (but of course there are no bugs, so...)

On Ubuntu, the following should do:

sudo apt install libeigen3-dev libgtest-dev libboost1.65-dev libglfw3-dev libomp-dev cmake

The usual cmake procedure

mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j

We make use of the Partio for saving particles to a file. To install the library, I would recommend creating a local install folder and installing it there. Roughly, the installation goes like this:

mkdir ~/local_build # Or anywhere else you would like to install the library.
export LOCAL_BUILD=~/local_build
git clone https://github.com/wdas/partio.git
cd partio
make -j prefix=$LOCAL_BUILD install

Create a build directory, run cmake to generate the makefile, run make, then run the binary.

mkdir build && cd build
cmake -DLOCAL_INCLUDE_DIR=$LOCAL_BUILD/include -DLOCAL_LIB_DIR=$LOCAL_BUILD/lib -DCMAKE_BUILD_TYPE=Release ..
make -j

From the build directory, type

./bin/mpm --scene ../scenes/snowman.toml

This should open a small 2D visualization of the simulation with a yellow snowman and start simulating.

Some command line options are available; these are the most important ones:

dt timestep (default 1e-4)

N number of MPM-grid cells per dimension (default 60)

particle-count number of particles per unit cube. Note: the simulation domain is a unit cube. (default 500000)

save-dir where to save the output (mesh or particles) (default is empty, so no output is generated)

laplacian smooth whether to apply laplacian smoothing to the output mesh (0 or 1, default is 0).

For further options, please refer directly to the file options.h.

Example usage:

./bin/mpm -N 80 --dt 1e-4 --save-dir ./output/ --laplacian_smooth 0

This project was created in collaboration Simon Duenser. The simulation implements the method as detailed in the paper

Hu, Yuanming, et al. "A moving least squares material point method with displacement discontinuity and two-way rigid body coupling." ACM Transactions on Graphics (TOG) 37.4 (2018): 1-14.

Their simplified implementation helped us get started.

The license is GPLv3 due to the marching cubes library used from libigl.