Research On Key Technology Of General Parallel Computing Framework In Multi-core Heterogeneous Environment

With the development of technology, especially the development of computertechnology, the amount of data in various industries grows exponentially, however,the traditional serial computing power cannot meet the growing demand for dataprocessing. In this context, parallel computing technology is advanced to quicklysolve large and complex computing problems. Parallel computing are not only closelyrelated to the technology and economic development of the country, but also directlyaffect the national defense capability and national security, such as nuclear explosionsimulation, complex systems accurate solver, genetic research, national confidentialcommunications encryption and decryption and so on. Parallel computing power is animportant symbol of national strength.Although parallel computing has been developed for many years and there aresome practical solutions and a lot of experience in some specific problems, thediversity of parallel computing is still less than the serial algorithm, and this study isnot yet a mature discipline. Corresponding to serial algorithm, Parallel Algorithms notonly need to consider the problem itself, but also need to consider appropriate parallelmodel, the processor architecture, network connection and other factors in order tomaximize efficiency, which must increase the difficulty of the parallel algorithmdesign and implementation.This thesis analyzes the various problems and research domestic and foreignrelated work on parallel computing, According to various problems of parallelcomputing model, A new multi-core layered heterogeneous general parallelcomputing is proposed to meet the computing needs of cluster of multi-coreprocessors, and some key technologies are studied in this paper..The detailed contents are as follows：(1) A new multi-core layered heterogeneous general parallel computingframework is proposed. Through the framework, the target problem is divided intoprogram model algorithm design, parallel program design and parallel programexecution. In program model algorithm design, developers design process modelalgorithm according to parameterized parallel machine; In parallel program design, developers utilize the parallel development platform to develop parallel programming;In parallel program execution, parallel programs is implemented on correspondingsoftware and hardware architecture, and instruction execution efficiency is optimizedby optimizing the calculation of parameter.(2) The framework is refined and implemented in this paper. According todifferent layers of layered heterogeneous model, the method library, parameter libraryand the program reuse library are designed to match the model. The design ofhierarchically parallel algorithm is carried out, and ultimately the parallel computingmodel could be dynamic, adaptive, reconfigurable and universal.(3) Between the different layers, language interpretation system and compilersystem are proposed to connect layers, which ensure the integrity and implementationof the system model.(4) The generic heterogeneous parallel computing framework is utilized todesign the parallel computing algorithm of prestack migration process. The prestackmigration program algorithm is the classic algorithm of oil exploration. The serialprogram of the algorithm is already complete, but its parallel algorithm has been amore complex issue. The framework well solutes its parallelism.The general parallel computing model provides a design language that is easy touse for application developers. The model achieves high-performance, accuracy anduniversality of parallel computing program design. So it has broad applicationprospects and significant social benefits. Layered heterogeneous parallel computingmodel provides a hardware-independent scalable programming interface for theapplication developers, establishes method library for some universal problem,establishes parameter library for the program running platform, and integratesdifferent computing resources, allocates computing tasks appropriately, and reducesthe difficulty of developing and workload of developers.