High-level FPGA Programming through Mapping Process Networks to FPGA Resources

We describe a simple and fast approach to FPGA programming that allows to efficiently exploit the numeric processing capabilities of recent FPGA chips. It basically consists in programming on top of a libraty of complex components for FPGA based scalable processor networks and providing a high-level programming intelface to it. The FPGA application is presented as a network of processes which is automatically transformed into a corresponding network of simple processor components by a compiler. The compiler then generates inidividual program code for each of the simple processors. The coarse-grained processor network is eventually compiled into an FPGA configuration bitstream using standard FPGA tools at close-to-interactive speeds. Our approach has the additional benefit of being fully compatible with processor programming and extendible to mixed multicomponent FPGA and processor systems. An experimental implementation of the process mapping scheme uses the p-Nets language that provides convenient structures for the presentation of the application processes and supports composite targets including processors linked to the FPGA chips. The evaluation of our concept on some FPGA chips includes an estimate of their floating point processing performances.