EXECUTION TIME ANALYSIS OF PARALLEL SCANNING AND SORTING FOR UNIFORM GRID CONSTRUCTION ON PC AND VR
The uniform grid is a commonly used technique for spatial partitioning across a variety of fields. Two parallel approaches, scanning and sorting, are typically used to construct the uniform grid by organizing spatial data into grid structures, with one method being selected for execution. While both methods can be executed in parallel to enhance performance, the application of parallel computing on virtual reality (VR) platforms remains less explored than on conventional desktop personal computers (PCs). To address this gap, this study evaluates the execution time of parallel scanning and sorting in uniform grid construction on both PC and VR platforms, aiming to explore the application of parallel computing in VR environments.
The impact of independent variables such as workgroup size, number of particles, and number of cells was investigated by testing multiple combinations of these parameters. Execution times for each combination on both platforms were measured using graphics processing units (GPU) timestamps to accurately capture GPU time.
Our results show that scanning consistently outperforms sorting on the PC, achieving an average speedup of 40.42%, with a median improvement of 41.67%. In contrast, no significant performance difference exists between the two algorithms on the VR platform, where the optimal algorithm depends on the specific parameter configuration.
These findings reveal several patterns that provide valuable insights for adjusting parameters in uniform grid construction on both platforms. The results also offer benchmarks and inform future optimizations for parallel scanning and sorting in uniform grid construction on both PC and VR platforms.
History
Degree Type
- Master of Science
Department
- Computer Graphics Technology
Campus location
- West Lafayette