Voxelizer implements an optimized version of the method described in M. Schwarz and HP Seidel's 2010 paper Fast Parallel Surface and Solid Voxelization on GPU's.

Dependencies The project has the following build dependencies:

• Nvidia Cuda 8.0 Toolkit (or higher) for CUDA + Thrust libraries (standard included)
• [oneAPI Toolkit] for running SYCL version on Intel GPU
• [Rocm Library] for running HIP version on AMD GP
• Trimesh2 for model importing. Latest version recommended.
• GLM for vector math. Any recent version will do.
• OpenMP

To build SYCL Version on Intel backend

mkdir build && cd build

CC=icpx CXX=icpx cmake ../ -DTrimesh2_INCLUDE_DIR="/path/to/trimesh2/include" -DTrimesh2_LINK_DIR="/path/to/trimesh2/lib.Linux64" -DGPU_AOT=PVC

make -sj

To run SYCL Version on Intel backend

ONEAPI_DEVICE_SELECTOR="level_zero:gpu" ./voxelizer_sycl -f ../../test_models/bunny.OBJ -s 1024 -i 20

To build SYCL Version on Nvidia backend

mkdir build && cd build

// For A100 CC=clang CXX=clang++ cmake ../ -DUSE_NVIDIA_BACKEND=YES -DTrimesh2_INCLUDE_DIR="/path/to/trimesh2/include" -DTrimesh2_LINK_DIR="/path/to/trimesh2/lib.Linux64" -DUSE_SM=80

// For H100 CC=clang CXX=clang++ cmake ../ -DUSE_NVIDIA_BACKEND=YES -DTrimesh2_INCLUDE_DIR="/path/to/trimesh2/include" -DTrimesh2_LINK_DIR="/path/to/trimesh2/lib.Linux64" -DUSE_SM=90

To run SYCL Version on Nvidia backend

./voxelizer_sycl -f ../../test_models/bunny.OBJ -s 1024 -i 20

To build SYCL Version on AMD backend

mkdir build && cd build

// For MI-100 CC=clang CXX=clang++ cmake ../ -DUSE_AMDHIP_BACKEND=gfx908 -DTrimesh2_INCLUDE_DIR="/path/to/trimesh2/include" -DTrimesh2_LINK_DIR="/path/to/trimesh2/lib.Linux64" -DUSE_SM=80

// For MI-250 CC=clang CXX=clang++ cmake ../ -DUSE_AMDHIP_BACKEND=gfx90a -DTrimesh2_INCLUDE_DIR="/path/to/trimesh2/include" -DTrimesh2_LINK_DIR="/path/to/trimesh2/lib.Linux64" -DUSE_SM=90

To run SYCL Version on AMD backend

ONEAPI_DEVICE_SELECTOR=hip:* ./voxelizer_sycl -f ../../test_models/bunny.OBJ -s 1024 -i 20

To build CUDA Version

mkdir build && cd build

cmake -DCMAKE_CUDA_COMPILER=/path/to/cuda/bin/nvcc -DTrimesh2_INCLUDE_DIR="/path/to/trimesh2/include" -DTrimesh2_LINK_DIR="/path/to/trimesh2/lib.Linux64" -DUSE_SM={compute} ../

make -sj

To run Cuda version

./voxelizer_cuda -f ../../test_models/bunny.OBJ -s 1024 -i 20

To build HIP Version

The default glm installation which is installed via "apt install libglm-dev" currently doesn't work for HIP version. Install the library from the source (https://github.com/g-truc/glm/tree/master) and point it that location while compilation.

To build and run HIP version please update these three files in Trimesh library. Comment "#pragma omp parallel for" in the below three files located under trimesh2/libsrc folder and rebuild the trimesh lib.

libsrc/TriMesh_connectivity.cc

libsrc/TriMesh_normals.cc

libsrc/TriMesh_pointareas.cc

mkdir build && cd build

HIP_PATH=/opt/rocm-5.4.3/hip CXX=/opt/rocm-5.4.3/bin/hipcc cmake ../ -DTrimesh2_INCLUDE_DIR=/path/to/trimesh2/include/ -DTrimesh2_LINK_DIR=/path/to/trimesh2/lib.Linux64 -DGLM_INCLUDE_DIR=/path/to/glm

make -sj

To run Hip version

./voxelizer_hip -f ../../test_models/bunny.OBJ -s 1024 -i 20

To view the output

To view the generated output file as a voxelized model use this online voxelizer, https://drububu.com/miscellaneous/voxelizer/?out=bin where you can import the file and view

@Voxelizer{cudavoxelizer17, author = "Jeroen Baert", title = "Cuda Voxelizer: A GPU-accelerated Mesh Voxelizer", howpublished = "\url{https://github.com/Forceflow/cuda_voxelizer}", year = "2017"}