Mali-T604 GPU的二维浮点矩阵运算并行优化方法

Parallel Optimization Method of 2D Floating-point Matrix Operation Based on Mali-T604 GPU

基于以三星的Exynos5250处理器为核心的Arndale Board嵌入式开发平台,对集成于处理器上的Mali-T604嵌入式GPU的GPGPU(General-Purpose computation on GPU)技术进行研究,并对不同运算规模的浮点矩阵乘法进行并行加速优化,提供实际测试结果。Linux操作系统上的实验结果显示,基于Mali GPU的并行浮点矩阵乘法方案相对原始串行算法而言,效率显著提高,并且运算规模的增大可以显著提高并行性。

The paper researches the applied method and practical effect of GPGPU technology based on Arndale Board platform which a-dopts Exynos5250 as core processor and designs the parallel optimizing solution for different scale of matrix multiplication problems based on GPGPU technology applied on Mali-T604 GPU which is integrated in Exynos5250 processor.The result of experiment on Linux OS shows that the computational efficiency of parallel matrix multiplication solution is improved significantly,and when the size of com-puting scale is vast,the improving effect is much more obvious.