Development And Research Of Improve Efficiency Of DVB Logo Inserter Through CUDA Technology

In the course of DVB technology popularization, customer requirement about high cost performance device increased. But the processing capacity of existing single DVB logo inserter is low, its cost performance can’t match the requirement of market. GPU has hundreds and even more processors, so it has powerful parallel processing ability. Use GPU can improve the processing capacity of single DVB logo inserter, improve its cost performance ratio, and improve its competitive forces.This topic introduced how to speed up the DVB logo inserter by CUDA technology through the GPU on the display adapter. For realize this purpose, need to translate a part of CPU’s code to GPU’s code, and make them work together with high performance. So this topic discussed the way define the scope of translation according the advantage difference between CPU and GPU. Introduced the way define the basic unit of GPU code deal with according the rule which balance the demand of real-time and the load demand of stream processor. Introduced the ways to define the interface between the CPU’s code and GPU’s code and the project’s module structure according the rule which balance efficiency and flexibility. Discussed the methods about solve the technologic problems such as compatibility of GPU compiler, memory pointer and limitation about code length. Talked about those methods used in debugging process, such as how to output log, debug with comparing, debug with multithread and deal with the exception of program. Talked about improve the performance by adjust work load and by change the mode of parameters transfer.Anyway, we can improve the system performance to more than 200% without hardware cost increment. By this way, we cut the cost of device and increase the competitive superiorities greatly.