No abstract

The paper suggests a technique for solving the matrix/set covering problem in all programmable systems-onchip. A novel very fast hardware accelerator is proposed and implemented in the programmable logic (PL) of a Xilinx Zynq microchip. The accelerator is managed by software running in the processing system (ARM Cortex-A9) available on the same microchip and communicating with the PL through high-speed interfaces. The results of implementation, experiments, and comparisons demonstrate significant speedup comparing to software running in general-purpose PC and in the ARM.

Many algorithms in informatics require a set of objects with similar properties to be grouped (clustered) on the basis of some predefined criteria. The proposed technique involves hierarchical merging in which software, responsible for solving the entire problem, is enhanced with highly parallel networks in hardware accelerators. Additional improvements are achieved with the aid of support methods that are sort and verification of object intersections that may also be autonomously used for other types of information processing and database management. It is shown and experimentally proved that the proposed solutions are efficient. They can be used in such areas as health care, statistical data manipulation and so on.

No abstract

The paper suggests a technique for extracting and filtering sorted subsets in a three-level computing system with such sub-systems as general-purpose computer (level 1), ARM Cortex-A9 (level 2), and reconfigurable logic (level 3). The last two levels are implemented in Zynq-7000 device available on the prototyping board ZC706. Communications between the levels 1 and 2-3 are organized through PCI express bus and interactions between components of levels 2 and 3through on-chip AXI interfaces. We studied two levels of software programs (running in PC and ARM), highperformance hardware accelerators implemented in Zynq-7000 programmable logic, and architecture enabling interactions and exchange of data between different levels. The selected for analysis sorting problem has high computational complexity and is widely required in data processing (data mining and statistical data manipulation, in particular). The results of experiments demonstrate that the elaborated architecture is efficient and permits fast solutions to be found. Proposals for potential further improvements are also given.