Differences between CUDA and OpenCL through a SAR focusing system

The aim of this work is to highlight the differences and the similitudes between two different platforms for GPUs’ coding, CUDA (Compute Unified Device Architecture) and OpenCL (Open Computing Language). The work will be explained further through a practical example, exposing how is organized the code of an actual SAR (Synthetic Aperture Radar) focusing system and showing the highlights and weaknesses of such system in comparison to a different coding of the same system. In conclusion, the main key points of the two different approaches applied to the SAR focusing system will be analyzed. Actually, both platforms are largely used and each of them has its highlights and weaknesses, so they are both used but for different aims and targets. The SAR focusing system which will be analyzed is actually written in CUDA, and an OpenCL code for it has not been written yet. In this work I will show why and when is better to choose one platform or the other, and I will show which solutions have to be made to porting the CUDA code to OpenCL code for the SAR focusing system. At first, I will talk in a general way about GPU, to better frame the topic of interest. Second, I will present in a deeper way the two different platforms, CUDA and OpenCL. Third, I will compare the two platforms, showing their highlight and weaknesses, and why and when is better choosing one or the other. Fourth, I will talk in a general way about the SAR focusing system, to better understand the taken example and in such a way to better explain the following step. Fifth, I will talk about how to porting the SAR focusing system’s CUDA code in OpenCL code and why doing it could be useful, and why not.