Reduced-Precision Acceleration of Radio-Astronomical Imaging on Reconfigurable Hardware

Stefano Corda,Akash Kumar,Roel Jordans,Albert-Jan Boonstra,Bram Veenboer,Henk Corporaal,John W Romein,Ahsan Javed Awan

doi:10.1109/access.2022.3150861

Abstract

Radio telescopes produce large volumes of data that need to be processed to obtain high-resolution sky images. This is a complex task that requires computing systems that provide both high performance and high energy efficiency. Hardware accelerators such as GPUs (Graphics Processing Units) and FPGAs (Field Programmable Gate Arrays) can provide these two features and are thus an appealing option for this application. Most HPC (High-Performance Computing) systems operate in double precision (64-bit) or in single precision (32-bit), and radio-astronomical imaging is no exception. With reduced precision computing, smaller data types (e.g., 16-bit) are used to improve energy efficiency and throughput performance in noise-tolerant applications. We demonstrate that reduced precision can also be used to produce high-quality sky images. To this end, we analyze the gridding component (Image-Domain Gridding) of the widely-used WSClean imaging application. Gridding is typically one of the most time-consuming steps in the imaging process and, therefore, an excellent candidate for acceleration. We identify the minimum required exponent and mantissa bits for a custom floating-point data type. Then, we propose the first custom floating-point accelerator on a Xilinx Alveo U50 FPGA using High-Level Synthesis. Our reduced-precision implementation improves the throughput and energy efficiency of respectively <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1.84\times$ </tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2.03\times$ </tex-math></inline-formula> compared to the single-precision floating-point baseline on the same FPGA. Our solution is also <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2.12\times$ </tex-math></inline-formula> faster and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$3.46\times$ </tex-math></inline-formula> more energy-efficient than an Intel i9 9900k CPU (Central Processing Unit) and manages to keep up in throughput with an AMD RX 550 GPU.

Highlights

T HE future generation of radio telescopes, such as the Square Kilometre Array (SKA) [1], will have to process a massive quantity of data using high-performance computing systems [2] with high energy efficiency [3]
We evaluate the performance of the radio-astronomical imaging algorithm on a Xilinx Alveo U50 FPGA employing HighLevel Synthesis and traditional floating-point data types
We evaluate the use of reduced-precision data types for the gridding kernel and make the following observations: Reduced precision applicability in radio-astronomical imaging: Different from artificial intelligence applications, where it is possible to highly reduce the data size, e.g. 1 or 8 bits [7, 101], radio-astronomical imaging needs higher precision for reconstructing sky images

Summary

Introduction

T HE future generation of radio telescopes, such as the Square Kilometre Array (SKA) [1], will have to process a massive quantity of data (in the order of TeraBytes per second) using high-performance computing systems (in the order of Exaflops per second) [2] with high energy efficiency [3]. The demanding data and computation requirements are mainly caused by the high-resolution images that must be processed to discover new objects in the sky such as stars, supernovas, galaxies, etc. The most dominant compute kernels of the radioastronomical imaging pipeline are the gridding, and degridding algorithms [5]. These kernels can be executed highly efficiently in single-precision floating-point accuracy using. Corda et al.: Reduced-Precision Acceleration of Radio-Astronomical Imaging on Reconfigurable Hardware

Methods

Results

Conclusion