Accelerating Legacy Applications with Spatial Computing Devices

TitleAccelerating Legacy Applications with Spatial Computing Devices
Publication TypeJournal Article
Year of PublicationSubmitted
AuthorsScionti A, Savio P, Vitali G, Vercellino C, Terzo O, Magarielli D, Spano E, Marconcini M, Poli F, Nguyen H-N
JournalJournal of Supercomputing
Abstract
Heterogeneous computing through acceleration fabrics 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 has grown, with Field Programmable Gate Arrays (FPGAs) as the most representative. While the combination of high performance, low power consumption and high adaptability constitutes an advantage, these devices still suffers from a weak software ecosystem, which forces application developers to use low level tools requiring deep knowledge of the underlying system, also leaving legacy code often unsupported, as the case of large body of Fortran applications. By realizing this, we describe a methodology for porting Fortran legacy code on modern FPGA architectures, preserving high performance/power ratios. As a test vehicle, we considered an industrial Computational Fluid Dynamics application, and
experimentally demonstrated that our methodology produces a synthesizable OpenCL-HDL code targeting Intel Arria10 and Stratix10 devices.