论文部分内容阅读
媒体处理领域中,大多数应用程序具有很高的计算需求,通用处理器难以提供满意的性能;专用处理器又存在成本较高、灵活性差等问题。于是斯坦福大学的imagine流处理器应运而生,在媒体处理领域有着出色的表现。科学计算领域也有很多的应用,它们具有很高的计算密集性和计算通信比,适合在流处理器上运行。因此斯坦福大学以imagine为基础,又开发了merrimac流处理器,并且提出采用merrimac构建P级超级计算机的想法。为了评估流处理器的性能、以及它在某特定领域的应用前景,本文选择了该领域常用的快速哈达玛变换作为benchmark,证实了流处理器性能上的优越,也指出了它在编程方面的一些不足。
In the field of media processing, most applications have high computational requirements, and general purpose processors are difficult to provide satisfactory performance. Dedicated processors have the problems of high cost and poor flexibility. So Stanford University imagine flow processors came into being, in the field of media processing has excellent performance. There are many applications in the field of scientific computing, they have a high computational density and computational communication ratio, suitable for running on stream processors. So on the basis of imagine, Stanford developed the merrimac stream processor and proposed the idea of using merrimac to build a P-class supercomputer. In order to evaluate the performance of the stream processor and its application prospects in a particular area, this paper selects the fast Hadamard transform commonly used in this field as the benchmark, confirms the superior performance of the stream processor, and points out its programming Some deficiencies.