PrestoZL: A GPU-accelerated High-throughput Jerk Search Toolkit for Binary Pulsars
Year: 2025
Type: article
Abstract: The Fourier domain jerk search algorithm, an integral component of the PRESTO software suite, has emerged as a key tool for detecting binary pulsars. However, it is a CPU-based jerk search and is a very computationally expensive process, particularly when exploring a broad range of search parameters. In order to address this challenge, we have developed PrestoZL, a GPU-accelerated, high-throughput jerk search toolkit. PrestoZL introduces an innovative GPU parallel design for the jerk search algorithm to mitigate performance degradation caused by memory-intensive operations. We have also developed a pipelined version of PrestoZL, which adds fine-grained orchestration to the CPU–GPU execution pipeline to alleviate the GPU stall problem during the search. The experiment conducted on a 30 minute observation using a machine equipped with an A100-40G GPU and 20 i7-12700K CPUs, shows that PrestoZL achieves an end-to-end speedup of 56.38× over a CPU-based jerk search in PRESTO with OpenMP. PrestoZL achieves search results that are fully identical to the CPU-based jerk search in PRESTO, including the number of detected pulsars, as well as the output search parameters and signal-to-noise ratio values.
Authors: Kuang Mao, Zhaorong Tang, Qiuhong Pan, Pei Wang, Huaxi Chen, Scott M. Ransom, Di Li, Xuefei Tang, Qi Wang, Yi Feng, Lei Chen, Donghui Quan, Zujie Ren
Language: English
Subfield: Astronomy and Astrophysics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $4,499
OpenAlex ID: W4413998528