Research On Task Partitioning And Mapping Of Coarse-grained Reconfigurable Cryptographic Logical Arrays

Coarse-grained reconfigurable processors have rich computing resources and interconnection resources,which can be applied to intensive computing,and have the characteristics of high energy efficiency and flexibility,so they have been widely concerned by scholars.Coarse-grained Reconfigurable Cipher Logical Array(CRCLA)is a kind of reconfigurable structure oriented to cryptographic algorithm.It adds a module suitable for cryptographic operation to the computing unit,and makes specific design in interconnection and control.However,this also makes the mapping method based on the traditional reconfigurable processor not fully applicable to the CRCLA structure.Therefore,This thisis analyzes the overall structure of CRCLA and the operational characteristics of cryptographic algorithms,and divides the mapping process into the compiler front-end and back-end.Aiming at automation generation,the front end abstracts the source code of cryptographic algorithm into a data flow diagram.Aiming at high energy efficiency,the backend divides the mapping problem of data flow graph on CRCLA into two parts: task partitioning and mapping,and carries out research respectively.The main research achievements of this thisis are as follows:1.In order to generate data flow diagram automatically,the front-end first carries on lexical and syntactic analysis to the source code to get syntax tree,and then carries on semantic analysis to generate data flow graph.Secondly,there is a problem in the source code that the function of finegrained operator combination is the same as that of CRCLA coarse-grained operator.Therefore,a graph embedding model based on attention mechanism is proposed for detection,recognition and replacement.Finally,optimization operations such as inline replacement,redundancy removal and parallelization are carried out to simplify the data flow graph.The experimental results show that compared with other compiler front-end,the data flow graph generated by this front-end is shorter and the mapping performance is improved by about 32%.2.Aiming at the task partition problem in the backend of CRCLA compilation,an ant colony task partition algorithm is proposed.Based on the cluster partition algorithm,the algorithm introduces the characteristics of ant colony algorithm to select nodes according to the probability of pheromone concentration matrix,which eliminates the shortcomings of mechanical selection nodes of cluster partition algorithm.In order to find the optimal solution,the pheromone increment formula and pheromone concentration update formula in the ant colony algorithm are improved to better optimize the pheromone concentration matrix.Experimental results show that compared with the typical task partitioning algorithm,the number of nodes and edges is reduced by 13.2% and 12.6% on average.3.Aiming at the mapping problem in the back-end of CRCLA compilation,an annealing genetic algorithm which can converge quickly to the global optimal solution is proposed by referring to simulated annealing algorithm and genetic algorithm.Firstly,the annealing algorithm is improved from the aspects of initial temperature and cooling coefficient,and secondly,the screening,crossover and mutation processes of the genetic algorithm are added to optimize the population.Meanwhile,in order to solve the difficulty of describing the process of mapping cryptographic algorithm to CRCLA,a description form of node cluster after partition is proposed,and a coordinate sequence description form is used to accurately display the resource occupation in CRCLA.The experimental results show that compared with the general mapping algorithm,the compilation time and performance are optimized by 30.6% and 23.4% on average.4.A CRCLA software development platform is designed.The software uses QT as the development platform to design a visual graphical interface,combined with the compilation of front-end and back-end partition and mapping research content,and finally exports CRCLA configuration information,reducing the complexity of developers mapping cryptographic algorithms.