arXiv subjects

Bruce Merry

Publications and source records attributed to Bruce Merry.

6 recordsLinked to original sources

arXiv2023

Efficient channelization on a Graphics Processing Unit

We present an implementation of a channelizer (F-engine) running on a Graphics Processing Unit (GPU). While not the first GPU implementation of a channelizer, we have put significant effort into optimizing the implementation. We are able to process four antennas each with 2 Gsample/s, 10-bit dual-polarized input and 8-bit output, on a single commodity GPU. This fully utilizes the available PCIe bandwidth of the GPU. The system is not as optimized for a single high-bandwidth antenna, but handles 6.2 Gsample/s, limited by single-core CPU performance.

astro-ph.IM↗

arXiv2022

The Astropy Project: Sustaining and Growing a Community-oriented Open-source Project and the Latest Major Release (v5.0) of the Core Package

The Astropy Project supports and fosters the development of open-source and openly-developed Python packages that provide commonly needed functionality to the astronomical community. A key element of the Astropy Project is the core package $\texttt{astropy}$, which serves as the foundation for more specialized projects and packages. In this article, we summarize key features in the core package as of the recent major release, version 5.0, and provide major updates for the Project. We then discuss supporting a broader ecosystem of interoperable packages, including connections with several astronomical observatories and missions. We also revisit the future outlook of the Astropy Project and the current status of Learn Astropy. We conclude by raising and discussing the current and future challenges facing the Project.

astro-ph.IM↗

arXiv2022

Tricolour: an optimized SumThreshold flagger for MeerKAT

We present Tricolour, a package for Radio Frequency Interference mitigation of wideband finely channelized MeerKAT correlation data. The MeerKAT passband is heavily affected by interference from satellite, mobile, aircraft and terrestrial transponders. Coupled with typical data rates in excess of 100 GiB/hr at 208kHz channelization resolution, mitigation poses a significant processing challenge. Our flagger is highly configurable, parallel and optimized, employing Dask and Numba technologies to implement the widely used SumThreshold and MAD interference detection algorithms. We find that typical 208kHz channelized datasets can be processed at rates in excess of 400 GiB/hr for a typical L-band flagging strategy on a modern dual-socket Intel Xeon server.

astro-ph.IM↗

arXiv2016

Faster GPU-based convolutional gridding via thread coarsening

Convolutional gridding is a processor-intensive step in interferometric imaging. While it is possible to use graphics processing units (GPUs) to accelerate this operation, existing methods use only a fraction of the available flops. We apply thread coarsening to improve the efficiency of an existing algorithm, and observe performance gains of up to $3.2\times$ for single-polarization gridding and $1.9\times$ for quad-polarization gridding on a GeForce GTX 980, and smaller but still significant gains on a Radeon R9 290X.

astro-ph.IM↗

arXiv2016

A Performance Comparison of Sort and Scan Libraries for GPUs

Sorting and scanning are two fundamental primitives for constructing highly parallel algorithms. A number of libraries now provide implementations of these primitives for GPUs, but there is relatively little information about the performance of these implementations. We benchmark seven libraries for 32-bit integer scan and sort, and sorting 32-bit values by 32-bit integer keys. We show that there is a large variation in performance between the libraries, and that no one library has both optimal performance and portability.

cs.DC↗

arXiv2015

Approximating W projection as a separable kernel

W projection is a commonly-used approach to allow interferometric imaging to be accelerated by Fast Fourier Transforms (FFTs), but it can require a huge amount of storage for convolution kernels. The kernels are not separable, but we show that they can be closely approximated by separable kernels. The error scales with the fourth power of the field of view, and so is small enough to be ignored at mid to high frequencies. We also show that hybrid imaging algorithms combining W projection with either faceting, snapshotting, or W stacking allow the error to be made arbitrarily small, making the approximation suitable even for high-resolution wide-field instruments.

astro-ph.IM↗