Implementation of parallel operations over streams in extensible processing platforms

The paper discusses the use of extensible processing platforms for the design of high-performance systems which combine fast parallel operations over data streams in programmable logic and problem-specific software running in advanced RISC machine. The streams contain information that needs to be analyzed and filtered. The main idea is to digitalize frequently changed data from numerous sensors and to represent each data item in form of a binary vector. It is shown that many analysis and filtering problems can be solved through Hamming weight counting for the vectors and comparison of the results with pre-given bounds (thresholds). The proposed architecture takes advantages from fixed plus variable computations and implements novel methods. The results of experiments and evaluations of the architecture in two Zynq-based prototyping boards are also presented.