Efficient Implementation of MS-OSM on GPU for Multiple DM Trials

Pulsars and Fast Radio Bursts (FRBs) are subject to dispersion effects caused by the interstellar medium (ISM), and coherent dedispersion is widely used to counteract this effect.However, when the dispersion measure (DM) is high, a large FFT length is required which will cause more computation complexity and resource requirement.This paper presents an efficient GPU implementation of MS-OSM tailored for multiple DM trials.This approach integrates multiple DM trials into a single kernel while leveraging GPU streams and concurrent CPU threads to maximize computational efficiency and memory utilization.It is demonstrated that MS-OSM significantly reduces GPU memory usage compared to traditional methods, enabling the simultaneous processing of more DM trials.Furthermore, the pipeline achieves high I/O throughput, making it a practical solution for real-time pulsar and FRB observations.Simulation results validate the accuracy and efficiency of our implementation, with successful dedispersion of real pulsar data and significant performance improvements over conventional methods.