Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Accelerating radio astronomy imaging with RICK: a step towards SKA-Mid and SKA-Low

Giovanni Lacopo, Emanuele De Rubeis, Claudio Gheller, Giuliano Taffoni, Luca Tornatore

TL;DR

RICK 2.0 tackles the data deluge of modern radio interferometry by delivering a portable, GPU-accelerated imaging pipeline built on HeFFTe for distributed FFTs. It replaces heavy all-to-all MPI communication with non-uniform domain decomposition, MPI-I/O, and HeFFTe-backed FFTs, achieving substantial performance gains on CPU and GPU hardware and mitigating prior bottlenecks where communication dominated runtime. The approach is validated with real MeerKAT and LOFAR data, showing strong scaling and dramatic speedups in gridding, FFT, and phase correction on GPUs, and demonstrated readiness for SKA precursors and the SKA era. The work highlights the broader impact of portable distributed FFTs for astrophysical imaging and outlines future avenues for asynchronous I/O to further improve time-to-solution.

Abstract

The data volumes generated by modern radio interferometers, such as the SKA precursors, present significant computational challenges for imaging pipelines. Addressing the need for high-performance, portable, and scalable software, we present RICK 2.0 (Radio Imaging Code Kernels). This work introduces a novel implementation that leverages the HeFFTe library for distributed Fast Fourier Transforms, ensuring portability across diverse HPC architectures, including multi-core CPUs and accelerators. We validate RICK's correctness and performance against real observational data from both MeerKAT and LOFAR. Our results demonstrate that the HeFFTe-based implementation offers substantial performance advantages, particularly when running on GPUs, and scales effectively with large pixel resolutions and a high number of frequency planes. This new architecture overcomes the critical scaling limitations identified in previous work (Paper II, Paper III), where communication overheads consumed up to 96% of the runtime due to the necessity of communicating the entire grid. This new RICK version drastically reduces this communication impact, representing a scalable and efficient imaging solution ready for the SKA era.

Accelerating radio astronomy imaging with RICK: a step towards SKA-Mid and SKA-Low

TL;DR

RICK 2.0 tackles the data deluge of modern radio interferometry by delivering a portable, GPU-accelerated imaging pipeline built on HeFFTe for distributed FFTs. It replaces heavy all-to-all MPI communication with non-uniform domain decomposition, MPI-I/O, and HeFFTe-backed FFTs, achieving substantial performance gains on CPU and GPU hardware and mitigating prior bottlenecks where communication dominated runtime. The approach is validated with real MeerKAT and LOFAR data, showing strong scaling and dramatic speedups in gridding, FFT, and phase correction on GPUs, and demonstrated readiness for SKA precursors and the SKA era. The work highlights the broader impact of portable distributed FFTs for astrophysical imaging and outlines future avenues for asynchronous I/O to further improve time-to-solution.

Abstract

The data volumes generated by modern radio interferometers, such as the SKA precursors, present significant computational challenges for imaging pipelines. Addressing the need for high-performance, portable, and scalable software, we present RICK 2.0 (Radio Imaging Code Kernels). This work introduces a novel implementation that leverages the HeFFTe library for distributed Fast Fourier Transforms, ensuring portability across diverse HPC architectures, including multi-core CPUs and accelerators. We validate RICK's correctness and performance against real observational data from both MeerKAT and LOFAR. Our results demonstrate that the HeFFTe-based implementation offers substantial performance advantages, particularly when running on GPUs, and scales effectively with large pixel resolutions and a high number of frequency planes. This new architecture overcomes the critical scaling limitations identified in previous work (Paper II, Paper III), where communication overheads consumed up to 96% of the runtime due to the necessity of communicating the entire grid. This new RICK version drastically reduces this communication impact, representing a scalable and efficient imaging solution ready for the SKA era.
Paper Structure (19 sections, 3 equations, 8 figures, 3 tables)

This paper contains 19 sections, 3 equations, 8 figures, 3 tables.

Table of Contents

  1. Introduction
  2. The $w$-stacking gridder
  3. Implementation
  4. Parallel I/O
  5. Domain Decomposition
  6. Gridding
  7. Fast Fourier Transform
  8. Methodology and RICK tests
  9. LEONARDO
  10. Booster Module
  11. Interconnect
  12. Compiler
  13. Small tests
  14. MeerKAT data
  15. Strong scaling tests
  16. ...and 4 more sections

Figures (8)

  • Figure 1: Schematic code architecture and workflow of RICK, based on the one in gheller2023 with the new steps that we ported on GPUs (reduce and FFT). Different kind of HPC enabling are highlighted with different colours.
  • Figure 2: Domain decomposition in rectangular slabs along the $v$ axis, with four MPI processes. Visibilities are originally distributed in time-order, whereas the grid is initialized in space-order, requiring an all-to-all communication once the gridding step is carried out.
  • Figure 3: New domain decomposition in rectangular slabs along the $v$ axis, with four MPI processes. This time the domain distribution is non even since due to the higher density of point in the central regions of the $u-v$ plane.
  • Figure 4: Dirty image produced with RICK starting from MeerKAT data for the Ophiuchus galaxy cluster botteon2025
  • Figure 5: Strong scaling of the different RICK components for the pure MPI configurations. Upper panel: strong scalability of core components. Total computed time is the whole core runtime from which I/O has been subtracted, communication time is the time for visibility redistribution and bucket sort time is the time which takes the visibility sorting algorithm along the $v$-axis. Lower panel: strong scalability of algorithmic components. Times for gridding, FFT and phase correction algorithmic steps.
  • ...and 3 more figures