The Research About The Parallel Programming Model Based On The Multi-core Embedded DSP

With the development of the high performance computer and the multi-core embedded processor, the highly efficient programming environment turns up and begins to develop rapidly, especially the progress of the parallel programming model. On account of the diversity of the parallel architecture, the exploitation of the parallel program becomes so difficulty. Hence, it’s the key issue how to simplify the development process of the parallel programming model. In the other hand, one of the focal points and hot problems worth studying is how to combine the multi-core embedded processors and the parallel programming models in order to take full advantage of the performance of the embedded processors.This dissertation makes a study on Phoenix based on TMS320C6678processor. First of all, this paper does a research on various communication mechanisms. The communication mechanism on C6678consists of the communication between the threads, the communication between the cores and the communication between the chips. Based on this, two methods to realize the Phoenix on C6678are proposed. One is on account of MessageQ and the share memory to establish the Phoenix on C6678. The communication and the synchronization are implemented by the message passing between cores. The other is in view of Multicore Navigator in C6678. QMSS and PKTDMA can be used to realize the communication and the synchronization with the help of the share memory. The last but not the least, there is an experimental analysis on this architecture.The three typical applications in Phoenix are word_count, linear_regression and histogram. The input test files of an arbitrary application are in three kinds of size. With the analysis of the statistical data of the executing time, it’s obvious that the executing time would reduce with the increase of the core number. Meanwhile, the greater the document processing is, the more obvious the decreasing trend would be. Besides, the method using Navigator based on hardware resource costs less time than the method based on MessageQ.