Novel method for solving inner product of dense matrix and sparse matrix through outer accumulation

Abstract

The invention belongs to the field of computer application equipment, and particularly relates to a novel method for solving an inner product of a dense matrix and a sparse matrix through outer accumulation. The method comprises the following effective steps: a, reading original data of a dense matrix A (MxK) and a sparse matrix B (KxN); b, performing row-direction data compression on the sparse matrix B (KxN), and performing row-direction compression on the sparse matrix B (KxN) in a two-dimensional layout into a row-direction compressed sparse matrix B (KxN) in a one-dimensional layout; c, reading the ith column of the row-direction compressed sparse matrix B (KxN) and the ith row of the dense matrix A (MxK), and performing outer product operation to generate ci; d, accumulating the obtained outer product results to obtain an inner product of the dense matrix A (MxK) and the sparse matrix B (KxN); wherein in the step b, aiming at the compression characteristic of the sparse matrix B (KxN), a first element value of a starting point of each row is reserved, whether the element value is zero or not is judged according to whether a non-starting-point element of each row is zero or not, whether the element value is zero or not is judged to be reserved or not, if the element value is zero, the element value is discarded, If the element value is non-zero, the element value is reserved.

Description

technical field [0001] The invention belongs to the field of computer application equipment, and in particular relates to a novel method for solving inner products of dense matrices and sparse matrices through accumulation and accumulation. Background technique [0002] Matrix inner product multiplication is the basis of algorithms in many fields such as numerical computing, machine learning, and engineering simulation. The inner product multiplication of dense matrix A (MxK) and sparse matrix B (KxN) is a common matrix inner product calculation scenario. [0003] The dense matrix A(MxK) and the sparse matrix B(KxN) are characterized in that the sparse matrix B(KxN) has a certain amount of zero elements. This part of zero elements can not participate in the calculation and still ensure the correctness of the calculation results. At the same time, if the zero elements are directly abandoned for storage, the memory utilization efficiency can also be improved. [0004] However...

AI Technical Summary

Problems solved by technology

However, in the existing outer accumulation calculation, the zero elements in the sparse matrix participate in the outer product calculation, which not only reduces the computational efficiency of the outer product operation but also increases the need to write the zero value to the memory after the outer product calculation, and then Reduced effective bandwidth for memory writes

Images