Parallel M2K data selection method based on graph theory and hardware space combination distribution

Abstract

The invention discloses a parallel M2K data selection method based on graph theory and hardware space combination distribution. The method comprises: dividing M pieces of data into L data boxes, and sorting the data therein; comparing the data of every two data boxes in all the current data boxes, extracting k top pieces of the data to store the same into an updated data box, and setting remainingone of the data boxes to an updated data box if the number of the current data boxes is an odd number; and if the number of the updated data boxes is one, determining that the data box is a requestedresult, and otherwise, returning to the previous step. According to the M2K data selection method disclosed by the invention, the M pieces of data are recombined through the L data boxes to reduce adata comparison range and reduce complexity; all data box updating processes can be simultaneously carried out, a parallelism degree and execution efficiency can be greatly improved, and time complexity can be optimized; and the same low hardware space complexity as existing serial methods can be realized, data carrying capacity and flexibility can be improved, and per-unit power consumption of hardware realization can be reduced.

Description

technical field [0001] The invention relates to a parallel M2K data selection method based on the combined distribution of graph theory and hardware space, and belongs to the technical field of digital integrated circuit algorithm optimization. Background technique [0002] With the development of big data and artificial intelligence, the algorithm has been transferred from the general-purpose computing core CPU or GPU to the customized hardware core. More and more multinational companies have accelerated the application of artificial intelligence algorithms through hardware FPGA or ASIC. Algorithms related to data processing cannot avoid the use of data selection and data calculation at the bottom layer. Data selection methods are widely used in intelligent algorithms, coding algorithms, etc., and their application proportion in a large number of data processing algorithms is higher. All of these make the data selection method more important in the application of hardware F...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to provide a parallel M2K data selection method based on the combined distribution of graph theory and hardware space, so as to solve the problems of existing data selection methods, such as high time complexity and space complexity, low execution efficiency, complex implementation, and low resources. Problems of high occupancy rate, poor flexibility, and high unit power consumption of hardware implementation

Images