Accelerating incoherent dedispersion

Abstract

Incoherent dedispersion is a computationally intensive problem that appears frequently in pulsar and transient astronomy. For current and future transient pipelines, dedispersion can dominate the total execution time, meaning its computational speed acts as a constraint on the quality and quantity of science results. It is thus critical that the algorithm be able to take advantage of trends in commodity computing hardware. With this goal in mind, we present analysis of the 'direct', 'tree' and 'sub-band' dedispersion algorithms with respect to their potential for efficient execution on modern graphics processing units (GPUs). We find all three to be excellent candidates, and proceed to describe implementations in C for CUDA using insight gained from the analysis. Using recent CPU and GPU hardware, the transition to the GPU provides a speed-up of 9x for the direct algorithm when compared to an optimised quad-core CPU code. For realistic recent survey parameters, these speeds are high enough that further optimisation is unnecessary to achieve real-time processing. Where further speed-ups are desirable, we find that the tree and sub-band algorithms are able to provide 3-7x better performance at the cost of certain smearing, memory consumption and development time trade-offs. We finish with a discussion of the implications of these results for future transient surveys. Our GPU dedispersion code is publicly available as a C library at: http://dedisp.googlecode.com/Comment: 15 pages, 4 figures, 2 tables, accepted for publication in MNRA

LOFAR Discovery of the Fastest-spinning Millisecond Pulsar in the Galactic Field

“…The complex voltage output for each tied-array beam-in the form of 200 Nyquist-sampled, dual-polarization subbands of 195 kHz each-was processed using GPU-accelerated software to perform coherent dedispersion and channelization with CDMT ) at steps of 1 pc cm −3 between dispersion measures (DMs) of 0.5 and 79.5 pc cm −3 . The resulting coherent filterbanks, sampled at 81.92 μs and 48.83 kHz in time and frequency, were dedispersed using the DEDISP library (Barsdell et al 2012). The dedispersed time series were searched for periodic signals using frequency domain acceleration searching (tools from PRESTO; Ransom 2001; Ransom et al 2002).…”

Pulsar–black hole binaries: prospects for new gravity tests with future radio telescopes

“…9 In Table 24 of Dewdney et al (2013), it is shown that the data processing for a pulsar survey with SKA1-mid is dominated by the acceleration search processing load of nearly 10 Peta operations per second; 10 times the processing power offered by the Einstein@Home network as of January 2013: http://einstein.phys.uwm.edu/ 10 Following the SKA1-mid survey parameters outlined in Table 24 of Dewdney et al (2013), and correcting for a jerk of ± 0.2 m s −3 (a value that could be observed in known compact significant progress in speeding up various aspects of pulsar search code has been made in recent years, primarily through the use of GPU technology (see e.g. Barsdell et al 2010;Magro et al 2011;Armour et al 2012;Barsdell et al 2012, Luo 2013 ). In the following sections our investigation is limited to searches utilizing the constant acceleration approximation only.…”

The Square Kilometer Array: cosmology, pulsars and other physics with the SKA

“…A few 10 4 trial of dispersion measures are required, and a computing power of 0.1 Petaflops. In addition, the de-dispersion is very I/O intensive (Barsdell et al 2012). …”