Sparse matrix acceleration calculation method, device, equipment and system thereof

Abstract

The invention discloses a sparse matrix acceleration calculation method, which comprises the following steps: in the operation process of a processor, receiving two sparse matrixes to be multiplied sent by a main memory; performing non-zero detection on each sparse matrix, and correspondingly storing the non-zero data in each sparse matrix and the row number and the column number where the non-zero data are located; Controlling non-zero data at corresponding row / column numbers in the two sparse matrixes to carry out product summation according to a matrix multiplication rule to obtain a product summation result; and storing the product summation result and the row number and the column number of the product summation result in the product matrix of the two sparse matrixes, returning the product summation result as product matrix data of the two sparse matrixes to the main memory, and enabling the processor to perform operation according to the product matrix data. When the sparse matrix is multiplied, only the non-zero data is calculated and stored, so that the occupation of the storage space is reduced, and the calculation speed is increased. The invention further discloses a device, equipment and a system based on the method.

Description

technical field [0001] The present invention relates to the field of sparse matrix technology, in particular to a sparse matrix acceleration calculation method, device, equipment and system thereof. Background technique [0002] With the near failure of Moore's Law, the computing performance of general-purpose processors (CPU) is also approaching the limit, but with the vigorous rise of cloud computing, big data, mobile Internet, artificial intelligence, and 5G, the computing requirements for processors are increasing. The higher it is, the contradiction between the two will be inevitable. [0003] Sparse matrices are generated in almost all large-scale scientific and engineering computing fields, including the current big data, image processing and other popular fields. Sparse matrices are data processing objects that often appear in the process of processor operations. Usually, the sparse matrices need to be processed by the processor. Multiplication processing, for examp...

AI Technical Summary

Problems solved by technology

[0005] Since the number of non-zero elements in a sparse matrix is ​​much smaller than the total number of matrix elements, and the distribution of non-zero elements is irregular, the calculation of the sparse matrix according to the conventional matrix calculation method is a great waste of storage space and extremely Greatly reduces the calculation speed

Images