A generic execution framework for shared FPGA-based accelerators

FPGAs are continuously increasing in both chip size and operating frequency. Dynamic reconfiguration is easier and more stable with current generation of hardware and software tools. These characteristics have made them more accessible to generic acceleration tasks instead of specialized functions. As a consequence, FPGAs are being deployed in more computing clusters than in the past. This leads to the need of a simple and reliable mechanism for sharing hardware resources between various software tasks. The current paper proposes a general method for accessing shared FPGA acceleration modules.