GPU支持的低延迟引力波数据处理

GPU accelerated low latency gravitational wave data processing

只有快速识别出引力波信号,才可以制导电磁望远镜及时探测到相关的电磁信号,对于全面了解引力波源所发生的天体物理过程,具有十分重要的科学意义.本文针对在激光干涉引力波天文台(Laser Interferometer Gravitational-Wave Observatory,LIGO)在线运行的并行求和无限冲击响应滤波(Summed Parallel Infinite Impulse Response,SPIIR)流水线数据处理系统,具体介绍了如何在图形处理单元(Graphics Processing Unit,GPU)上采用多种性能优化手段,大幅度提高其数据处理的速度来满足LIGO低延迟实时数据处理的要求.并进一步描述了如何在高新LIGO(Advanced LIGO)基础上实现极低延迟数据处理的计算方法、性能优化方法与相关的工具支持.

To enable prompt electromagnetic （EM） follow-up observations of GW events, it is very critical to identify the gravitational wave signal in real time so we can investigate the astrophysical events completely. In this paper we first described the LIGO （Laser Interferometer Gravitational-Wave Observatory） SPIIR （Summed Parallel Infinite Impulse Response） online data processing pipeline which can reduce the latency to meet real time requirement on GPU （Graphics Processing Unit） with multiple optimization methods. Moreover, to achieve extremely low latency for advanced LIGO data processing, this paper also presents the proposed computational method, performance optimization method and visualization tool.