| With the increasing development of multi-core processor hardware, the most needed for multi-core processors are support from system software and application development. To acquire best performance of the software based on multi-core processors, such software should be improved to take advantage of multi-core processors. Otherwise, the efficiency of these software would be less than that of single-core CPU, not mention to utilize the powerful calculating capability of multi-core processors. Based on these, how to design and develop software based on multi-core processors, and to increase the performance of system by take advantage of multi-core processors have become the most important problem during software development of nowadays.The main topics of this thesis are as following:(1).study the development of multi-core processers and multi-core technology around the world, analyze the key problems of multi-core parallel programming and the difference between multi-core programming and single-core programming. Also, multi-thread programming, communication and synchronization of thread are discussed, the multi-core parallel programming pattern for real use is deeply researched.(2).Analyze the characteristics of digital image process and total framework to solve problems in digital image process, such as too much calculation, slowly operation, and so on. Find out bottleneck and hotspot of digital image process applications, propose improvable part by using multi-thread through taking advantage of Intel(?) VTune and Intel(?) Parallel Studio. Use OpenMP to develop program for such bottleneck and hotspot area.(3).Make parallel optimize design in three aspects(cycling parallel and data protection, dynamic thread number setting, thread manage policy), propose a method, which is highly extendable, to debug and analyze multi-core parallel applications.(4). Count the accelerate ratio in dual-core and quad-core situation by calculating serial and parallel runtime. Experiment proves that the implementation of optimization saves 40%-60%time when compared to the original serial program, increases the acceleration process of applications. The advantage of combing the multi-core platforms, parallel programming theory and development is verified, achieving the desired optimization goal. |
PDF Full Text Request