Fast Radius Search Exploiting Ray Tracing Frameworks

Source Code for "Fast Radius Search Exploiting Ray Tracing Frameworks".

Dependencies

• tinyobjloader

Requirements

• Visual Studio 2019
• NVIDIA CUDA v10.1 (or higher)
• NVIDIA OptiX SDK v7.2.0 (or higher)

Project Files

The main project files are:

• main.cpp - Invokes a set of test cases with the point clouds demonstrated in the paper
• radius_search.hpp/cpp - Initializes, builds and executes radius search methods using the OptiX API
• radius_search.cu - Implements all the relevant kernels described in the paper for radius and truncated knn searches

Additionally, the project uses some code that has been retrieved/modified from the NVIDIA OptiX framework and is located in the sutil folder.

How To Build

• Add an environment variable CUDA_PATH_V10_1 pointing to the NVIDIA CUDA installation directory, e.g. C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.1/. Ideally, this variable is already set during the CUDA installation
• Add an environment variable OPTIX_PATH_V7_2 pointing to the NVIDIA OptiX installation directory, e.g. C:/ProgramData/NVIDIA Corporation/OptiX SDK 7.2.0/
• Edit the compile_optix.bat to set the corresponding CCBIN path for the C++ compiler
• Edit the compile_optix.bat to set the sm_arch variable that best suites the underlying device
• Compile assets/radius_search.cu to assets/ptx/radius_search.ptx by executing the compile_optix.bat through the command line
• Run Visual Studio and compile the project

How To Run

Run the optix radius search.exe that is located in the x64\Release or x64\Debug folder, depending on which configuration the project has been compiled with.

The expected command line output is :

• The Acceleration Data Structure (ADS) size
• The construction time of the ADS given the sample set
• The total time to execute the queries for a specific search method invocation

How to invoke custom code

Setting up a custom routine will require to :

• Create an OptiX context
• Create an instance of bvh_index and initialize it
• Build the ADS based on the AABBs constructed from the samples

The available radius search functions are :

• radius_search_count which returns a per query counting of neighbor samples based on the given radius along with some global statistics
• radius_search which returns per query index and distance buffers to samples (invokes radius_search_count internally to guarantee the gathering of every neighbor sample)
• truncated_knn which returns per query index and distance buffers to samples given a user predefined maximum capacity per query

All of the above searching invocations also implement a brute force approach that simply triggers a ray generation kernel that loops over all the available samples without invoking any BVH traversal for reference purposes.

Direct modification of radius_search.cu is required if a custom intersection function is required during the gathering stage.

Future TODO

• Implement Progressive Photon Mapping as demonstrated in the paper