Enhanced rolling cache architecture with prefetch

Abstract

Most of image processing algorithms such as filters are sequentially processed. So, data reusability should be considered to maximize processing efficiency. Because of its high performance and resource requirement, data refill by block based cache architecture has been proposed to maximize data reusability for a portable device image processing. However, if non-continuous data access happens, block based cache architecture degrade performance because data reusability is broken. In this paper, prefetch based enhanced rolling cache method is proposed. Proposed method not only improves data reusability for sequential processing cases, but also maintains efficiently even if data reusability is broken. Our experimental results show that the proposed method reduces CPU stall time and the number of cache refill clocks counts up to 60% compared to previous methods.