Research And Implementation Of Data Parallel Programming Platform Based On Multi-Core

With the development of multi-core processor, the application put forward higher requirement to performance. However, because of the processing capability of every core of multi-core processor is usually lower to former single core processor, using multi-core does not meaning high performance, the increasing of performance needs to application with parallelization, especially to the application with large information such as image processing application, using parallel computing to acquire high performance is a main method.Most of applications used for image processing are designed for specific hardware, and these applications are too closely associated with the hardware, so the applications developed for specific hardware can not be used in other hardware with different architecture. To solve this problem, this paper provides and designs a data-parallel programming platform, which face to the application of image processing, provide programming service interface with three kinds of programming services contain parallel programming language, programming service API and parallel library of image processing. Providing this programming platform aims to keep the programmer away from the details of hardware, developed parallel program with high performance, portability and generalization only to concern about the algorithm.The parallel language interface of this paper is abstract designed according to 2 kinds of typical data parallel language, which is scalable and can be used not only the hardware platform of the experiment in this paper but also other SIMD hardware platform. Data parallel technology is used to accelerate in all the hardware the platform supported, such as accelerating by the feature of hardware and parallel programming in SIMD and accelerating by SSE instruction of Intel in PC.The application developed on the platform verified that the data parallel programming platform in this paper is usable, can be used easily and reliable, and be accelerated in all experiment hardware environments.