Enabling seamless execution on hybrid CPU/FPGA systems: Challenges & directions

Today's computing systems are heterogeneous, with diverse micro-architectures. It is common to design systems comprising multi-core CPUs, Digital Signal Processors (DSP) and Graphic Processing Units (GPUs). In addition to these devices, the extensive configurability and parallelism in Field Programmable Gate Arrays (FPGAs) has proven to be advantageous for accelerating complex computational problems. In comparison to CPUs, DSPs and GPUs, FPGAs have a distinctly different microarchitecture. In this context, enabling seamless application execution in a system composed of FPGAs alongside CPUs, is a major challenge. As a consequence, simplifying programmability of hybrid CPU/FPGA systems demands innovations in system software support. There has been research in extending the traditional CPU-only system techniques to CPU/FPGA based hybrid systems to make way for FPGA-based mainstream computing. This article surveys techniques in dynamic task management aimed at reducing or completely eliminating the burden of operating system writers for CPU/FPGA hybrid systems and provide insight to researchers to explore further.