A Comparative Performance Analysis of Convolution W/O OpenCL on a Standalone System

Initial approach of this paper is to provide a deep understanding of OpenCL architecture. Secondly, it proposes an implementation of a matrix and image convolution implemented in C (Serial Programming) and OpenCL (Parallel Programming), to describe detailed OpenCL programming flow and to provide a comparative performance analysis. The implementation is being carried on AMD A10 APU and various algebraic scenarios are created, to observe the performance improvement achieved on a single system when using Parallel Programming. In the related works authors have worked on AMDAPPSDK samples such as N-body & SimpleGL to understand the concept of vector data types in OpenCL and OpenCL-GL interoperability, have also implemented 3-D particle bouncing concept in OpenCL & 3D-Mesh rendering using OpenCL. Lastly, authors have also illuminated about their future work, where they intend to implement a novel algorithm for mesh segmentation using OpenCL, for which they have tried to form a strong knowledge base through this work.