Chinese Journal of Computers   Full Text
Title	The Optimization Approaches of Organizing Streams on Imagine Processor
Authors	YANG Xue-Jun1) ZENG Li-Fang2) DENG Yu1) TANG Yu-Hua1)
Address	1)(National Key Laboratory of Parallel and Distributed Processing, School of Computer Science, National University of Defense Technology, Changsha 410073) 2)(Jiangnan Remote Sensing Institute, Shanghai 200072)
Year	2008
Issue	No.7(1092—1100)
Abstract &   Background	Abstract Due to the characteristics of stream applications and the insufficiency of conventional processors when running stream programs, stream processors which support data-level parallelism become the research hotspot. This paper presents two means, stream partition (SP) and stream compression (SC), to optimize streams on Imagine. The results of simulation show that SP and SC can make stream applications take full advantage of the parallel clusters, pipelines and three-level memory hierarchy of the Imagine processor, and then reduce the execution time of stream programs. Keywords imagine stream processor; stream application; stream optimization; stream partition; stream compression Background This paper addresses the problem of optimizing the performance of programs on the stream processor, which is an emerging architecture that addresses the problem of memory wall. As the stream processor architecture exposes more fine-granularity parallel mechanisms and memory operations to the programmers, the performance is heavily determined by the programmers. The previous works on this domain mainly focus on the organization of streams from arrays and the optimizations of kernels. The most testing programs are from media applications. This paper provides a method to improve the utility of the hardware of the stream processor, i.e. the usage of stream buffers from another point of view. Further more, it provides a method to reduce the memory transfers according to the characteristics of the applications. The two methods are effective to both media and scientific applications. This work is supported by NSFC projects: Key Technologies of Peta-flops High Performance Computing (60621003) and High Productive Parallel Computer Architecture (60633050). High performance computing is a foundational, foresighted and strategic crucial technology for solving significant application problem of nation and national defense construction. These projects revolve the long term demand of high performance computing from the national significant application domain and hold the significant transformation opportunity of the international high performance computing domain technology route. In this direction, this research group has carried out a 64-bit stream processor FT64, which is the first implementation for scientific applications. A lot of papers have been presented in ISCA07, ICPP07, ISPA06&07, ACSAC06&07 and so on. This paper explores the optimization methods on the stream processor through programming and gives the indications for the design of compiler.