OpenCL을 이용한 병렬 LDPC 디코딩 성능분석

Abstract

Many mobile devices have multi-core architectures as application processor (AP). A processor has been required to provide various services on the mobile. Processor engineers increased the number of cores to solve these needs. Mobile device has a limited battery capacity, so the increased number of cores decreases the life time of battery. Low-power core and high-performance core are integrated into a heterogeneous multi-core architecture, and this heterogeneous architecture is receiving attention as the next generation mobile AP because these kinds of heterogeneous multi-core satisfy user requirements like high performance and low power. Recent research has focused on the performance evaluation of parallel algorithm on a homogeneous multi-core. In this paper, a novel Low Density Parity Check (LDPC) decoding method using OpenCL is introduced. We used OpenCL framework for a parallel software implementation of LDPC for China Multimedia Mobile Broadcasting (CMMB) standard while existing research used the Nvidia CUDA framework on Nvidia hardware. CUDA framework is available only for Nvidia devices; however, the OpenCL framework can be used not only for various vendors’ CPU such as Intel and ARM, but also for GPU like nVidia and ATI. We measured the running time of LDPC decoder parallelized with these frameworks In case of SNR=4, each running time of LDPC decoding implementations by C and CUDA was 675.3 sec and 2.8 sec. We showed that the performance of CUDA using GPU is faster than C using single core. We parallelized the LDPC decoder using OpenCL framework, and we measured the performance of parallelized LDPC decoder using OpenCL. it has a strong portability and it is an open standard that can be used for any devices of various vendors