Research On Key Optimization Technology Of Image Programming Based On SIMD Platform Adaptive Programming Model

The widespread use of high-performance computers has led to a dramatic increase in demand for efficient parallel software.In the face of the huge,complex and demanding task of timeliness,how to make full use of the hardware and software resources of high-performance computers to speed up the calculation task quickly;in the face of accumulated valid and very efficient serial Applications,how to migrate them to supercomputers,these problems have long been the most concern and headache in high performance computing.Serialization of serial programs has become an important research direction in the field of computer research.In addition,since the 1960 s digital image technology,image processing in various fields are widely used,such as space science research,medical imaging,photography and so on.With the continuous integration of digital image field and computer intelligent computing technology,how to convert the serial image processing program into a higher performance parallel image processing program has become a hotspot in computer research.Parallelized compilation technology is one of the main ways to solve serialization of serial programs.The so-called parallel compiler technology refers to a variety of technical means to automatically detect the serial program in the potential parallelism,and the serial program into the equivalent of efficient parallel program translation technology.Parallelized compilation studies have begun with the emergence of high-performance computers,and the parallelization studies have made gratifying progress from existing research results.However,so far,the parallelization of the compilation system has a distance from the actual engineering application,There is still much to be done in the field.The ultimate goal of parallelism is to generate efficient parallel programs that conform to the characteristics of the target architecture.Therefore,this paper realizes a parallel optimization method for image processing program for a specific application domain-image processing,with an efficient SIMD platform adaptive programming model-PARAC framework.The working process of PARAC framework is divided into three parts: first,some of the calculation process of serial image processing algorithm is abstracted,parac language is used to describe it,and PARAC framework is used to optimize parac program.Finally,we get the SIMD architecture A high-performance parallel C program.In this paper,by reading the relevant literature and information,the research status of SIMD programming and the related extended research are obtained,and the basic research contents and common methods of SIMD programming model are summarized.Through the longitudinal analysis and comparison of some image processing algorithms,it is found that the Laplacian pyramid algorithm and the image sharpening algorithm are representative,so the Laplacian pyramid algorithm and the image sharpening algorithm are used as the research object,PARAC framework in the above two algorithms in the serial program parallelization used in the optimization method.The main contents of this paper are as follows:Firstly,introduce the background and research status of SIMD programming.Secondly,introduce the basic framework of SIMD programming model-PARAC,including the provision of compiler support;Secondly,the specific work of this paper is elaborated.First,introduce the ideas and processes of the Laplacian pyramid algorithm and image sharpening algorithm.Second,in the PARAC framework,from the perspective of SIMD program on the algorithm of the functional analysis and module division,and the optimization process to explain the difficulties.Thirdly,it focuses on the parallel optimization method used in the PARAC framework,including the design of the optimization strategy and the concrete realization scheme.Finally,the correctness of the parallel program of the image processing algorithm generated by the PARAC framework is verified by experiments,and it is compared with its serial program,which directly reflects the performance benefit of the PARAC framework automatically generated parallel program.