| Title |
A Case Study: Practical Parallelization of BRDF Estimations in CUDA |
| Authors |
이혁(Hyuck Yi) ; 백선호(Sunho Baek) ; 홍혜란(Hyeran Hong) ; 김준성(JunSeong Kim) |
| DOI |
https://doi.org/10.5573/ieie.2024.61.3.49 |
| Keywords |
Parallelization; Performance; GPU; CUDA; BRDF |
| Abstract |
With advancements in computer hardware, there is a growing interest in parallel programming and its widening scope of applications. In particular, parallel programming using GPUs has become essential in fields demanding extensive high-speed computations, thanks to lower entry barriers in terms of system configuration cost and implementation complexity. However, careless parallelization that overlooks the specific characteristics of GPU's hardware makes the efforts of engineers meaningless and fails to fully exploit the hardware's potential. In this paper, we present a practical parallelization approach for general engineers with a BRDF estimation algorithm. We use CUDA to parallelize the BRDF algorithm at various levels and scrutinize several hardware metrics to ensure efficient resource utilization. By taking into account both algorithm and hardware characteristics, we can enhance the performance of parallel programming. Experiments conducted with real-world data demonstrate that efficient computing is possible with little effort in parallel programming. The observed speedup ranges from 4.93 to 64.10 depending on the chosen parallelization approach and problem size. The parallelization process presented in this paper will be helpful to researchers who want to adopt parallel programming for the first time as well as those who have already adopted it and aim for higher performance. |