Research And Realization Of Automatic And Optimal Transformation Of C Language For Heterogeneous Reconfigurable Processor

The general purpose processor cannot meet the need of the new computing intensive applications,and the design speed of application specific integrated circuits cannot keep up with the update speed of complex applications.So the reconfigurable processors have become a hot research topic for their flexibility and high ability.In this paper,we study on the coarse grained heterogeneous reconfigurable processor called GReP.It includes a main control processor and several RPU(reconfigurable processor unit),each of which contains a co-processor and 4 PEA reconfigurable computing arrays.In order to fully exploit its parallelism,we designed a corresponding complier and a high-level language based on it.For facilitating the verification of the functions of GReP and do transplantation of traditional C language applications,it is important to design an automatic translation and parallelization system for the compiler.Oriented to the compiler for reconfigurable processor GReP,this paper simplifies the corresponding high-level language GR-C and adds an automatic transform module,which transforms the C language code to the the GR-C high-level language code,as the pretreatment end of the compiler.And the system also increased the appropriate loop optimization module,so that the automatic generated GR-C code can be more suitable for the compiler back-end,which can fully exploit the parallel characteristics of the hardware architecture.The downward compatibility of the automatic transform system is strong,so it can rapidly update according to the changes of hardware architectures or GR-C language.In this paper,we use the automatic transform module to achieve several algorithm of EEMBC,which is industry-standard benchmarks for embedded systems,and the compiler generates the configuration information.Data shows that the GR-C code,which is generated from the automatic transform system,is able to run on GReP correctly.Compared with general processor,it can speed up algorithms by 2.0?5.6.Compared with the manual configuration,it has similar acceleration effect.