Accelerating geographic processing using GPUs: Implemented in OpenCL

Abstract

In This thesis, we present how a geographical process can be increased in execution time and asymptotic complexity, by moving the processing algorithm from the Central Processing unit (CPU) to the Graphics Processing Unit (GPU). We also investigate different memory strategies on the GPU in order to further decrease the execution time for the algorithm. To improve the asymptotic complexity two new algorithms are investigated and implemented, the first algorithm is based on the concept of separability, and the second algorithm is a state-of-the-art algorithm called Gaussian filter kernel. The outcome of our work is an approach of how algorithms on the CPU that has great potential of parallelism can be moved to the GPU to improve the execution time. In order to evaluate the different algorithms, tests regarding the execution time and outcome accuracy were conducted. Lastly, we concluded the overall success of the improvement regarding the execution time and reduction for the asymptotic complexity.