基于传输触发架构的图像降晰专用处理器设计

DESIGNING A TRANSPORT TRIGGERED ARCHITECTURE-BASED EXCLUSIVE PROCESSOR FOR IMAGE BLURRING

针对天文图像处理过程中的空间变换核降晰算法,提出一种基于传输触发架构(TTA)的专用处理器设计,以完成算法的数据密集型运算,并在基于C*CORE340处理器的电子系统级平台之上对该处理器进行验证。TTA的细粒度并行数据传输使该处理器获得更高性能,其设计的特殊功能单元能加快计算,且计算流程和数据存储方式能减少数据传输的频率,提高了计算速度。实验结果表明,对于19×19的内核和2K×2K的图像帧,该处理器架构的速度比基于PC的纯软件实现提高了14.7倍。

In this paper we propose an exclusive processor design which is based on transport triggered architecture （TTA） aiming at the blurring algorithm using space-varying kernel in the process of astronomic image treatment for completing the data-intensive computing of the algorithm. The processor is verified on the electronic system-level platform using C ＊ CORE340. This processor can get higher performance with the help of the fine-grained parallel data transport in TYA. Special function units are designed to speed up the computation. The computation flow and the way of data storage are designed to reduce the frequency of data transmission and to improve the computation speed. Experimental results demonstrate that for the kernel size 19 × 19 and image frame size 2K×2K, the proposed processor architecture can achieve 14.7 times improvement in speed compared with the software-only implementation based on PC platform.