关键词: 低密度奇偶校验码, 图形处理器, 统一计算设备架构, 并行计算, 共享内存, 寄存器

Abstract:

To take advantage of the high speed parallel feature of the GPU and the parallel section in the regular LDPC codes decoding process, a method is proposed by which the GPU is used to accelerate decoding of regular LDPC codes. In this method, edges of nodes are used in parallel decoding instead of nodes themselves to improve the utilization of threads. At the same time, the use of the high-speed on-chip GPU memory-shared memory and registers to store data makes data reduce dependence on global memory and shorten access time. Simulation results show that, by using parallel computing on edges and the on-chip memory, the decoding speed can be 5.32 ～ 10.41 times relative to the LDPC codes decoding program that does not use the optimization method of this paper based on the GPU.

Key words: low density parity-check(LDPC) codes, graphic processing unit(GPU), compute unified device architecture(CUDA), parallel computing, shared memory, registers