Research On Parallel Accelerating Algorithm Based On OpenCL And Realization On FPGA

With the continuous progress of the science and technology of digital information,parallel computing has been penetrated in many fields of high-performance development,The speed and performance of single-platform or single computing is not completelyconform with increase of the demands of people. The platform of high performance andlow power consumption has become the development trend of the future. From singlecore to multi-core processors, from CPU, GPU, DSP to FPGA, and then to multi-corecombination between them, even to the GPU heterogeneous platform based on OpenCL.Although they have realized the performance improvement of the parallel computing, allof them can’t solve the problem that how to collaborative design hardware and software.The Altera SDK for OpenCL combines the strong parallel architecture of FPGA andOpenCL parallel programming model, it can achieve the powerful system and obtainhigh performance in the ultra low power consumption. Using this SDK, the systemdevelopers and programmers who Familiar with the C language can easily and quicklyDevelop the application of high performance and efficiency based on FPGA.This thesis firstly studies the relevant knowledge of parallel heterogeneous, andsummarizes the research status of both here and abroad and Current problems; Thenmake a detailed analysis of the OpenCL parallel architecture, and for the characteristicsof the implemented on CPU, GPU and FPGA platform made a detail comparison.Then the thesis introduces the advantages of achieving OpenCL and the methodand the architecture to achieve in FPGA, and introduces detailly the developmentenvironment and implement method of developing OpenCL in the DE5-NET FPGAplatform.Finally, the thesis takes respectively sparse Matrix-Vector Multiplication(SpMV)and Sobel image edge detection as the experimental object, not only analysis theparallelism in algorithm, but also accelerates the kernel algorithm in the DE5-NET FPGAplatform with OpenCL, and compares with the basic GPU platform, performance testing,analysis and compares the acceleration effect and power consumption. Experimental results show that compared with the serial algorithm of traditionalCPU and the parallel algorithm of design based on GPU platform with OpenCLacceleration. The parallel algorithms of OpenCL based on FPGA platform designed inthis thesis can ensure the realization of the algorithm in quality and efficiency,effectively improve the efficiency that allows computing speed has been significantlyaccelerated, and in a great extent, reduce the power consumption of the system.