基于PAAG的图形图像算法的并行实现

Parallel Implementation of Graphics Rendering and Image Processing Algorithm Based on PAAG

为了解决当前的CMOS技术遇到"功耗墙"和"散热墙"等问题导致的很难通过提高主频来提升芯片性能的问题,文中提出了一种新型多态同构阵列处理器—PAAG(Polymorphic Array Architecture for Graphics)。该阵列机在一个芯片上集成了多个处理器,能够通过将各种高性能复杂的算法合理分解映射到该平台上实现并行计算。通过结合使用数据并行、操作并行的计算方法,对固定渲染管线的图形算法以及由国际标准组织Khronos提出的计算视觉标准Open VX1.0中的Kernel函数图像算法进行了深入分析,并给出了基于这些算法在PAAG上的并行化设计。通过在PAAG硬件平台对应的仿真环境上进行各个算法的并行实现,得到了算法在多个处理单元上的运行时钟,由此计算出算法在多个处理单元上运行的加速比。实验结果表明,文中的并行化设计方法在PAAG上能够实现对图形图像算法的线性加速,与串行相比,效率更高。

In order to solve the problem that current CMOS technology has already met the ＂wall＂ of power and cooling which may cause the issue of improving the performance by improving the frequency of the chips, present a new polymorphic isomorphic array processor, called PAAG （ Polymorphic Array Architecture for Graphics and image processing）. This array integrates multiple processing elements on a chip,it can realize parallel computing of the high-performance and complex algorithms by dividing and mapping them to the platform. By combining the data-level and the operation-level parallel calculation methods,the algorithms of the fixed rending pipeline and these of OpenVX1. 0, a standard of computer vision, proposed by the international standard organization Khronos, are in-depth analyzed in this paper. And the parallel design of these algorithms are proposed based on PAAG. The soft simulation platform of PAAG can give the result number of the running clock of the parallel implementation of the algorithms. Then through calculating the speed-up ratio, can conclude that the parallel implementation of the design method and its experimental results show that the algorithms of both graphic and image can be parallel acceleration on PAAG platform, compared with the serial processing, this method can be more effective.