Method and circuit for executing matrix multiplication and SOC

Abstract

The invention discloses a method for executing matrix multiplication, and belongs to the field of data processing. The method comprises the following steps: acquiring a matrix A1, a matrix B2 and an index matrix; generating n matrixes A2 according to the index matrix and the matrix A1, wherein the n matrixes A2 are in one-to-one correspondence with n columns of the index matrix in sequence, t columns of each matrix A2 are in one-to-one correspondence with t elements of one column in the corresponding index matrix in sequence, and each column of each matrix A2 is a column indicated by one element in the corresponding index matrix in the matrix A1; and generating a matrix C according to the n matrixes A2 and B2, wherein n columns of the matrix C are in one-to-one correspondence with the n matrixes A2 in sequence, n columns of the matrix C are in one-to-one correspondence with n columns of the matrix B2 in sequence, and each column of the matrix C is the product of one corresponding matrix A2 and one corresponding column of the matrix B2. The matrix multiplication operation can be completed only by reading the data from the matrix A1 once, and the maximization of data reusability is realized.

Description

technical field [0001] The present application relates to the field of data processing, in particular to a method, circuit and system on chip (System on Chip, SOC) for performing matrix multiplication. Background technique [0002] Artificial intelligence technology is widely used in terminals, edge side, cloud, etc., to realize image recognition, target detection, voice translation and other functions. Artificial intelligence technology is often realized through deep learning network. For deep learning networks, operators that affect performance and have a large amount of calculation, such as convolution and inner product (inner product), account for up to 99% of the calculation. These operators can be expanded into matrix multiplication Matrix operations. Of course, matrix is ​​a common form of data expression, and the operation of matrix multiplication is often applied in other fields. [0003] At present, when performing matrix multiplication operations, the operation ...

AI Technical Summary

Problems solved by technology

[0005] The present application provides a method, circuit and SOC for performing matrix multiplication operations, which can solve the problems of low operation efficiency of matrix multiplication operations, address conflicts when reading data, and low data reusability in related technologies

Images