基于CPU_MIC_GPU异构架构的Roberts算法优化探究

在高分辨率图像日益普及的情况下,Roberts边缘检测的处理速度急需进一步提高。在CPU表现不尽如人意的情况下,基于CPU/GPU和CPU/MIC的高度并行运算的研究愈加深入。在分析Roberts算法特点的基础上,将能并行的部分移植到GPU和MIC上进行。完成基于CPU/GPU和CPU/MIC的异构架构上的Roberts算法实现,并针对CPU/MIC上将程序进行向量化优化。实验结果表明,在相同单精度浮点运算能力下,GPU处理低分辨率图像的速度更快、加速比更高,但处理高分辨率图像时MIC的加速比最高为23.52,高于GPU的21.43。

The Research of Optimized Roberts Algorithm Based on Heterogeneous CPU/MIC/GPU Architecture

In the case of the growing popularity of high-resolution images,the Roberts edge detection processing speed urgent need to further improve. Because the disappointing performance of CPU, the height parallel computing research based on the CPU/GPU and CPU/MIC become deepened. On the basis of analysis of the characteristics of Roberts algorithm, the part of the parallel will be respectively transplanted into GPU and MIC. Accomplish the Roberts algorithm which is based on CPU/GPU and CPU/MIC in a Heterogeneous architecture, and optimization the procedure aim at CPU/MIC. Experimental results show that, in the same single-precision FLOPS, the GPU processing low resolution image than faster, higher, but the MIC when dealing with high resolution images faster, speed is higher.