动态任务分配cuda线程束步进体绘制 volume rendering using dynamic cuda warp marching.pdfVIP

第28 卷 第 10 期 计算机辅助设计与图形学学报 Vol. 28 No.10 2016 年 10 月 Journal of Computer-Aided Design Computer Graphics Oct. 2016 动态任务分配CUDA 线程束步进体绘制 1) 2)* 3) 2) 2) 孙万捷 , 高 瞻 , 潘海燕 , 王杰华 , 蒋峥峥 1) (南通大学电子信息学院 南通 226019) 2) (南通大学计算机科学与技术学院 南通 226019) 3) (南通大学附属医院心血管内科 南通 226001) (gaozhan@) : 针对标准 CUDA 光线投射体绘制过程中因线程束内线程计算量不均产生线程束分化, 导致计算资源利用率 低的问题, 提出 CUDA 线程束步进的算法. 首先分析标准 CUDA 实现导致线程束分化的原因, 提出将光线积分映射 至线程束上, 线程束内所有线程同步分段积分直至光线终止, 以避免线程束分化; 然后结合光线积分的数学原理和 GPU 的硬件特性提出线程束内光线积分的算法; 最后针对静态线程束任务分配方式导致负载失衡的缺点, 提出动态 线程束任务分配的实现算法. 实验结果表明, 动态任务分配线程束步进算法的性能较标准CUDA 实现可获得 1.9~7.9 倍的加速效果. ：CUDA; 线程束; 体绘制; 资源利用率 ：TP391.41 Volume Rendering Using Dynamic CUDA Warp Marching Sun Wanjie1), Gao Zhan2)*, Pan Haiyan3), Wang Jiehua2), and Jiang Zhengzheng2) 1) (College of Electronic Information, Nantong University, Nantong 226019) 2) (College of Computer Science and Technology, Nantong University, Nantong 226019) 3) (Cardiovascular Medicine, Affiliated Hospital of Nantong University, Nantong 226001) Abstract: A CUDA warp marching method for ray casting volume rendering is proposed to address the problem of low computational resource utilization resulted from warp divergence due to irregular workload for each thread in a warp. We firstly analyzed the reasons for warp divergence in standard CUDA implementation. Warp diver- gences are eliminated by integrating each single ray with all the threads in a warp, which executes instructions in lock-step. The algorithm of integrating one single ray within one warp is introduced after detailing the mathe- matical principles and GPU hardware specifications. Dynamic work scheduling strategy is also