Sample Code for "Accelerating Force-Directed Graph Drawing with RT Cores"

This project contains sample code for the paper

S. Zellmann, M. Weier, I. Wald, "Accelerating Force-Directed Graph Drawing with RT Cores", IEEE Visualization 2020, Shortpapers.

Preprint (researchgate link)

Source Code Overview

While some of the source code files merely contain "boilerplate code", the following implementation files contain the actual graph drawing code:

Implementation of the graph drawing framework: gd.h / gd.cu These files implement the various phases of the graph drawing algorithm with CUDA. For the repulsive phase, the files contain a simple CUDA kernel implementing the naive method, or alternatively call into the OWL/OptiX or LBVH nearest neighbor programs and kernels that are implemented elsewhere.

Implementation of the OptiX device programs: optixSpringEmbedder.h / optixSpringEmbedder.cu These files contain the optix intersection and bounds programs that implement the ray tracing-based nearest neighbor query. For the OWL/host side, see the routines in gd.cu that call into these programs.

Implementation of the LBVH data structure: lbvh.h / lbvh.cu These files contain the implementation of our reference method based on the fast BVH tree construction algorithm by Karras.

Implementation of the user interface: main.cpp This file implements the user interface. The graph drawing algorithm runs in a separate thread from the user interface. We use a simple OpenGL renderer that only draws the graph edges as GL lines. This is not how we generated the images for the paper; for that, we rather use the software Tulip. Our graph layouts can be exported to the format supported by Tulip.

Dependencies

The sample code includes the following third-party libraries as submodules:

• OWL wrappers library on top of OptiX 7
• Visionaray ray tracing library, but only used for GUI and linear algebra
• CmdLine library to handle command line arguments (archived, use CmdLine2 instead)

In addition, the following dependencies should be installed by the user:

• C++11 compliant compiler, tested on Ubuntu 18.04 (g++ (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0)
• CMake
• OpenGL
• GLEW
• NVIDIA CUDA Toolkit
• Optix 7
• Boost
• GLUT or FreeGLUT

On Ubuntu 18.04, the dependencies can, e.g., be installed using the following command: sudo apt-get install cmake libglew-dev libboost-all-dev freeglut3-dev. CUDA and OptiX should be installed via the links provided above and according to the respective instructions. Make sure that you have a proprietary NVIDIA graphics driver (comes, e.g., with the CUDA toolkit) installed and loaded by the kernel.

Building

The following assumes that the OptiX and CUDA base directories can be found via environment variables; otherwise, adjust the paths accordingly:

git clone --recursive https://github.com/owl-project/owl-graph-drawing

cd owl-graph-drawing
mkdir build
cd build

cmake .. -DCMAKE_BUILD_TYPE=Release -DOptiX_INCLUDE:PATH=${OptiX_DIR}/include -DBIN2C=${CUDA_DIR}/bin/bin2c
make

Running

After successfully building the application, the application binary will be located in the build/ directory and is called gd. By default, gd loads an artificial graph; you can start the iterative graph drawing algorithm by pressing Key-SPACE. gd can also load .csv files exported with Gephi. Artificial data sets initialized upon application start can also be customized via the command line. For detailed usage info, type ./gd -help:

Usage:
   ./gd [OPTIONS] [files...]

Positional options:
   [files...]             A list of input files

Options:
   -C[=<int>]             Clusters
   -bench[=<bool>]        Benchmarking mode to disable keypress event handling
   -bgcolor               Background color
   -camera=<ARG>          Text file with camera parameters
   -connected[=<bool>]    Generate connected graph
   -dt[=<ARG>]            Select data generation mode:
      =artificial         - Artificial Graph Generation
      =tree               - Tree Generation
      =file               - File Input
   -epc[=<int>]           Edges per Clusters
   -fullscreen            Full screen window
   -height=<ARG>          Window height
   -mode[=<ARG>]          Select graph layout mode:
      =naive              - Naive Implementation
      =rtx                - RTX Mode
      =lbvh               - LBVH Mode
   -n[=<int>]             Maximum number of iterations
   -npc[=<int>]           Nodes per Clusters
   -o[=<string>]          Output tlp file
   -r[=<int>]             Number of repetitions when loading data
   -refit_after[=<ARG>]   Refit RTX BVH after N iterations
   -trDegree[=<int>]      Tree data generation degree
   -trDepth[=<int>]       Tree data generation depth
   -width=<ARG>           Window width

Use the following mouse controls to navigate the 3D graph:

• LMB: Pan the scene.
• MMB: Pan the scene.
• RMB: Zoom into the scene.

Citation

If you want to refer to this project in your own scientific work, please cite the following research paper:

@INPROCEEDINGS{zellmann:2020,
  author={Zellmann, Stefan and Weier, Martin and Wald, Ingo},
  booktitle={2020 IEEE Visualization Conference (VIS)},
  title={Accelerating Force-Directed Graph Drawing with RT Cores},
  year={2020},
  volume={},
  number={},
  pages={1-5},
}

License

This sample code is licensed under the MIT License (MIT)