Research On High Level Synthesis Optimization Methods Based On Genetic Algorithm

With the rapid development of integrated circuits,embedded FPGA not only has low power consumption,high performance and reconfigured,can support the use of any manufacturing process and controls the size of any application.Embedded FPGA IP technology is convenient for developers to explore their architecture and design space of high-level synthesis.After the emergence of artificial intelligence,convolution neural network instantly becomes the focus of academia and industry,however,due to the large amount of calculation,not on an ordinary computer or embedded devices for real-time processing.In addition,on the basis of the existing algorithm written in C/C + +,such as stereo matching algorithm,DCT and FIR,etc.,the algorithm has very wide application field,especially in the field of digital processing(e.g.,communication or image processing),but there is also the problem of a large amount of computation.Therefore,it is necessary to propose a system architecture based on high-level synthesis for search optimization strategy.So that the various algorithms in different fields can be automatically integrated into hardware description language or IP core,which realizes the hardware acceleration of the algorithm.This paper design a system architecture based on genetic algorithm for high level synthesis optimization strategy search by studying the current situation of domestic and overseas and analyzing the optimization strategy of various core algorithms under HLS.Furthermore,this paper proposes the affine partition algorithm is applied to the HLS dependence analysis,has realized the automatic generation of TCL file,extract the useful information of data report and the automatic optimization strategy,design of fitness function and genetic operators.Finally,this paper proves the effectiveness of the system through experimental analysis and comparison.