Content area
Ray tracing is a process used commonly in computer graphics and in physics to track light photons and particles, respectively. Much research was found on improving execution times for the computer graphics applications; however, in the time frame of this literary review, almost no research was found on improving the execution times for the physics applications that were relevant to this problem. Two ray trace algorithms, a STL raytrace and a conebeam raytrace, were optimized using OpenMP and CUDA.
The STL raytrace algorithm achieved up to a 13.41x speedup using OpenMP. CUDA was not used for this algorithm due to the large amounts of memory allocations required in the algorithm. The conebeam raytrace used both OpenMP and CUDA along with knowledge of branches in coding to improve the execution time. The algorithm was able to improve the runtime by 14.3x using the most optimized OpenMP. There was an unexpected result using CUDA, in which the first calculation of the executable file took around 200ms extra to run, causing different runtimes for the first calculation in the executable file and each calculation after. Using the most optimized CUDA code, the algorithm achieved a 13.87x speedup on the first calculation and an 18.5x speedup for each calculation after. The CUDA optimization also allowed for answers an average of 8.5% more accurate than the original sequential code when comparing the computed results to the actual results.
Both algorithms achieved an improvement in execution time using the different methods. Each algorithm was unique in the way they needed to be parallelized and tested. These differences help broaden the knowledge on how the execution times of physics ray trace algorithms can be improved.