Heterogeneous Computing with OpenCL

Chapter 7. OpenCL Case Study

Convolution

This chapter discusses the implementation of more advanced optimizations of OpenCL kernels to improve the performance of a convolution filter.

Keywords Convolution, example program, OpenCL

Introduction

In Chapter 4, we introduced a basic convolution example using OpenCL images. Images provided the benefit of automatically handling out-of-bounds accesses (by clamping or wrapping accesses), which simplified the coding that would have been required for the cases in which the convolution filter accessed data outside of the image. Thus, image support may reduce control flow overhead and provide caching and data access transformations that improve memory system performance. When targeting GPUs, the automatic caching ...