Accelerating Legacy Applications with Spatial Computing Devices

TitleAccelerating Legacy Applications with Spatial Computing Devices
Publication TypeJournal Article
Year of Publication2023
AuthorsSavio P, Scionti A, Vitali G, Viviani P, Vercellino C, Terzo O, Nguyen H-N, Magarielli D, Spano E, Marconcini M, Poli F
JournalJournal of Supercomputing
Date Published11/2022
ISSN Number1573-0484
Accession NumberWOS:000890149800002
Other NumbersScopus 2-s2.0-85142931345
KeywordsFPGA, High-performance computing, Spatial computing
Heterogeneous computing is the major driving factor in designing new energy-efficient high-performance computing systems. Despite the broad adoption of GPUs and other specialized architectures, the interest in spatial architectures like fieldprogrammable gate arrays (FPGAs) has grown. While combining high performance, low power consumption and high adaptability constitute an advantage, these devices still suffer from a weak software ecosystem, which forces application developers to use tools requiring deep knowledge of the underlying system, often leaving legacy code (e.g., Fortran applications) unsupported. By realizing this, we describe a methodology for porting Fortran (legacy) code on modern FPGA architectures, with the target of preserving performance/power ratios. Aimed as an experience report, we considered an industrial computational fluid dynamics application to demonstrate
that our methodology produces synthesizable OpenCL codes targeting Intel Arria10 and Stratix10 devices. Although performance gain is not far beyond that of the original CPU code (we obtained a relative speedup of × 0.59 and × 0.63, respectively, for a single optimized main kernel, while only on the Stratix10 we achieved × 2.56 by replicating the main optimized kernel 4 times), our results are quite encouraging to drawn the path for further investigations. This paper also reports some major criticalities in porting Fortran code on FPGA architectures.
Refereed DesignationRefereed