Sphere Tracing - ASL Project

By Felix Sarnthein, Irfan Bunjaku, Lukas Heimes, and Patrick Ziegler

Project Work for MSc Advanced Systems Lab course offered at ETH Zurich (263-0007-00L). For details on implementation, algorithm, optimizations and evaluation check out 03_report.pdf.

Compiling

This project uses the standard cmake build process:

mkdir build
cd build
cmake ..
make -j

Several options can be set for the cmake command:

• -DINSTRUMENT=ON: Turns on instrumentation to count flops
• -DSANITIZE=ON: Turns on the address and undefined behavior sanitizer for debugging.
• -DCMAKE_BUILD_TYPE=Release/-DCMAKE_BUILD_TYPE=Debug Use release mode or debug mode.

On macOS:

• -DCMAKE_C_COMPILER=/usr/local/bin/gcc-10 and -DCMAKE_CXX_COMPILER=/usr/local/bin/g++-10: Actually use gcc instead of Apple clang. gcc-10 can be installed using Homebrew.

For benchmarking

cmake -DCMAKE_BUILD_TYPE=Release -DINSTRUMENT=OFF -DSANITIZE=OFF

should be used and for regular development

cmake -DCMAKE_BUILD_TYPE=Debug -DINSTRUMENT=OFF -DSANITIZE=OFF

The sanitizers should only be turned on for finding bugs and as a sanity check (to see if it can detect undefined behavior) on the code because it dramatically increases compile and runtime.

Running

Compiling will produce an executable called main in the build directory. It takes four mandatory arguments and two optional ones:

./main <impl> <input> <width> <height> [<output>] [<reference>]

The first one is the name of the implementation to run. Each impl_* folder in src is one such implementation, the name for it is everything after impl_.

The second parameter is the path to a JSON file containing a scene definition followed by the desired width and height of the image in pixels.

The fifth argument is optional and denotes the path where the resulting ppm (or pfm) image file should be written (any existing file will be overwritten). If the argument is not specified, no image will be written. The last optional argument is the path to the reference ppm image (used for output validation).

Create Benchmark scenes

Running the Python scripts will create benchmark scenes with random objects. create_scene.py creates n scenes of type shape

python create_scene.py <shape> <n> <output>

create_scene_all_shapes.py creates n shapes of each shape type.

python create_scene_all_shapes.py <n> <output>