Speaker
Piyush Panchal
Description
The SMART Pulsar survey has generated petabytes of raw data, of which only a small fraction has undergone coarse processing for pulsar searches using PRESTO, a toolkit that supports basic thread- and node-level parallelism. To search a larger portion of the dataset and enable more refined searches within reasonable timeframes, software improvements are necessary. The bulk of computation in pulsar searches is dominated by de-dispersion and jerk search algorithms, which follow a hit-and-trial approach and are naturally parallelizable. In this work, we present our contributions to GPU-based implementations of these two algorithms and compare their performance with the standard PRESTO library.