DescriptionFPGAs play a critical part in heterogeneous compute platforms as flexible, reprogrammable, multi-function accelerators. They enable hardware performance with the programmability of software. The industry trend towards software-defined hardware challenges not just the traditional architectures - compute, memory, network resources - but also the programming model of heterogeneous compute platforms.
Traditionally, the FPGA programming model is narrowly tailored and hardware-centric. As FPGAs become part of heterogeneous compute platforms and users expect the hardware to be “software-defined”, they must be accessible not just by hardware developers but by software developers which require the programming model of FPGAs to evolve dramatically. This presentation focuses on outlining a highly evolved, software-centric programming model which will enable FPGAs for software developers through a comprehensive solutions stack including FPGA-optimized libraries, compilers, tools, frameworks, SDK integration, and an FPGA-enabled ecosystem. We’ll also show a real-world example using Machine Learning Inference acceleration on FPGAs.