Software acceleration on Xilinx FPGAs using OmpSs@FPGA ecosystem

Abstract

The OmpSs@FPGA programming model allows offloading application functionality to Xilinx Field Programmable Gate Arrays (FPGAs). The OmpSs compiler splits the code (written in C/C++ high level language) in two parts, targeting the host and the FPGA. The first is usually compiled by the GNU Compiler Collection (GCC), while the latter is given to the Xilinx Vivado HLS tool (hereafter HLS) for high level synthesis to VHDL and bitstream used to program the FPGA. OmpSs@FPGA is based on compiler directives, which allow the programmer to annotate the part of the code to automatically exploit all Symmetric MultiProcessor system (smp) and FPGA resource available in the execution platform.

This technical report provides both descriptive and hands-on introductions to build application-specific FPGA systems using the high-level OmpSs@FPGA tool. The goal is to give the reader a baseline view of the process of creating an optimized hardware design annotating C-based code with HLS directives. We assume the reader has a working knowledge of C/C++, and familiarity with basic computer architecture concepts (e.g. speedup, parallelism, pipelining).