Computing Sorted Subsets for Data Processing in Communicating Software/Hardware Control Systems

Abstract

Computing and filtering sorted subsets are frequently required in statistical data manipulation and control applications. The main objective is to extract subsets from large data sets in accordance with some criteria, for example, with the maximum and/or the minimum values in the entire set or within the predefined constraints. The paper suggests a new computation method enabling the indicated above problem to be solved in all programmable systems-on-chip from the Xilinx Zynq family that combine a dual-core Cortex-A9 processing unit and programmable logic linked by high-performance interfaces. The method involves highly parallel sorting networks and run-time filtering. The computations are done in communicating software, running in the processing unit, and hardware, implemented in the programmable logic. Practical applications of the proposed technique are also shown. The results of implementation and experiments clearly demonstrate significant speed-up of the developed software/hardware system comparing to alternative software implementations.

Hardware accelerators for solving computationally intensive problems over binary vectors and matrices

“…Note that there are some other operations that are not as frequent as the operations listed above. Let us consider four examples from [11,15,17,19]. In [11] the described above operations help malicious attacks to be detected in data transmission.…”

“…In [19] HW computations allow well-organized objects satisfying some requirements to be extracted from unstructured and semi-structured data. Solving this and similar problems might be helpful in the scope of big data [20].…”

Hardware accelerators for processing clusters in binary vectors

“…Let us consider an example. Suppose the set D = {10, 0, 21,41,89,7,21,22,14,10,21,7,10,21,24, 28} of integers is given. Sorting permits the following sorted set {0, 7,7,10,10,10,14,21,21,21,21,22,24,28, 41, 89} to be produced.…”

“…Suppose the set D = {10, 0, 21,41,89,7,21,22,14,10,21,7,10,21,24, 28} of integers is given. Sorting permits the following sorted set {0, 7,7,10,10,10,14,21,21,21,21,22,24,28, 41, 89} to be produced. In the next step, all pairs of adjacent items are compared, and the vector V is built: V = (010110011100000), where the left 0 is the result of the comparison of 0 and 7, the next 1 is the result of the comparison of 7 and 7, the next 0 is the result of comparison of 7 and 10, etc.…”

“…Therefore, the most frequent item appears four times (the computed Hamming weight plus 1) in the original set and this is item 21 (the item whose index is 7 in the sorted set). Other practical applications are described in [16][17][18][19][20][21][22][23][24]. Since many systems have hard real-time constraints, performance is important and hardware accelerators may provide significant assistance.…”