Accelerating Pulsar Searches in PRESTO via Parallelization 80 views

Searching for pulsars is a computationally expensive process: radio telescopes take data at high time and frequency resolution, across large bandwidths, and we search over a large parameter space that includes dispersion measure and acceleration and jerk due to binary motion. However, there are opportunities to speed up the process by executing searches in parallel. A commonly used software package, PRESTO, implements parallel searches through OpenMP, an API that allows tasks to be spread across CPUs. By modifying the implementation of OpenMP and introducing vectorized operations in complex multiplication, we demonstrate improvements in acceleration search speed using multiple CPUs. We present a parallel version of rfifind, the routine that detects and flags radio frequency interference (RFI). After applying thread-safe replacements for statistical routines and altering how interference candidates are recorded, we show that performance closely matches ideal predictions from Amdahl's Law for large and noisy datasets.

BS (Bachelor of Science)

pulsars; high-performance computing; parallelization

English

2026-05-10