Research On FPGA-based Heterogeneous Computing Platform And Kernel Programming Design

In recent years,with the rapid development of computing-intensive fields,single-core and multi-core processors can't meet people's demand for computing power in the face of explosive growth in computing demand,while heterogeneous computing shows significant advantages.In the meanwhile,FPGA has been used more and more widely in highperformance computing with its excellent energy efficiency ratio and computing capabilities.Therefore,heterogeneous computing combined with CPU+FPGA has gained widespread attention.However,it is mainly provided by major chip manufacturers,which are limited to their respective products,without supporting external expansion.In order to solve the problem of poor portability of the CPU+FPGA heterogeneous system,we relied on the Open CL standard and explored the design of a heterogeneous computing hardware framework,which supports the Open CL protocol.While SRIO is responsible for the communication between the host and the devices,the execution kernel on the hardware can be developed using HLS.And we summarized relevant development suggestions for the HLS development of the kernel.The main work of the thesis includes:(1)We bulit a set of hardware architecture supporting Open CL basic functions and implemented on FPGA based on Open CL,which includes SRIO communication system,simple instruction set,and logic modules for the execution of the kernel programs.With this system,the development of kernel functions can not only be obtained with the use of IP generated by HLS.Besides,the hardware architecture using SRIO communication can support the expansion of FPGA devices.(2)A variety of optimization suggestions are proposed for the designed hardware architecture.Including the optimization of data transmission,storage access and data processing.(3)Linear discriminant analysis,which was developed by HLS,was used for system testing.Several HLS efficient development methods and kernel interface adaptation schemes were summarized during the development process.In the end,the functions of the hardware system and its support for Open CL are verified through the execution of the algorithm under the hardware framework with host.