GPU-optimised low-latency online search for gravitational waves from binary coalescences

Xiaoyang Guo, Qi Chu, Zhihui Du, Linqing Wen

Research output: Chapter in Book/Conference paperConference paper


Low-latency detection of gravitational waves (GWs) from compact stellar mergers is crucial to enable prompt followup electro-magnetic (EM) observations, as to probe different aspects of the merging process. The GW signal detection involves large computational efforts to search over the merger parameter space and Graphics Processing Unit (GPU) can play an important role to parallel the process. In this paper, Summed Parallel Infinite Impulse Response (SPIIR) GW detection pipeline is further optimized using recent GPU techniques to improve its throughput and reduce its latency. Two main computational bottlenecks have been studied: the SPIIR filtering and the coherent post-processing which combines multiple GW detector outputs. In the filtering part, inefficient memory access is accelerated by exploiting temporal locality of input data, where the performance over previous implementation is improved by a factor of 2.5-3.5x on different GPUs. The post-processing part is improved by employing multiple strategies and a speedup of 12-25x is achieved. Once again, it is shown that GPUs can be very useful to tackle computational challenges in GW detection.

Original languageEnglish
Title of host publication2018 26th European Signal Processing Conference, EUSIPCO 2018
PublisherEuropean Signal Processing Conference, EUSIPCO
Number of pages5
ISBN (Electronic)9789082797015
Publication statusPublished - 29 Nov 2018
Externally publishedYes
Event26th European Signal Processing Conference, EUSIPCO 2018 - Rome, Italy
Duration: 3 Sep 20187 Sep 2018


Conference26th European Signal Processing Conference, EUSIPCO 2018

Fingerprint Dive into the research topics of 'GPU-optimised low-latency online search for gravitational waves from binary coalescences'. Together they form a unique fingerprint.

Cite this