Symmetric matrix and vector multiplication parallel computing method and system

Abstract

The invention provides a symmetric matrix and vector multiplication parallel computing method and system, and the method comprises the steps: obtaining a symmetric matrix and a vector, and determining a thread count; dividing sub-regions; in each sub-region, defining a first axis and / or a second axis parallel to the symmetry axis; calculating a symmetry axis region and a vector, and writing into a register; performing two times of different first multiplication calculation on the first axis to obtain two groups of different data, and respectively writing the two groups of different data into a register; and performing two different second multiplication calculation on the second axis to obtain two or four groups of different data, and respectively writing the data into the register to obtain a calculation result of multiplication of the symmetric matrix and the vector. According to the scheme, multiplication calculation of the symmetric matrix and the vector can be accurately and rapidly achieved, the processing sequence of the elements in the matrix is reasonably planned, resources are reasonably allocated to all the threads, write-in conflicts can be effectively avoided, multi-thread parallel calculation is achieved, the calculation efficiency is high, loads of all the threads are balanced, and thread resources are reasonably distributed.

Description

technical field [0001] The invention relates to the field of matrix data processing, in particular to a parallel computing method and system for multiplying a symmetric matrix and a vector. Background technique [0002] A real number matrix A is a data block composed of M×N real numbers. And a vector x can be considered as a matrix composed of N×1 real numbers. The multiplication result y between the matrix A and the vector x is an M×1 vector. The multiplication of matrices and vectors is defined as follows: [0003] [0004] The parallel calculation of matrix multiplication by the processor can effectively shorten the data processing time. When developing parallel algorithms, an important issue is how to avoid write conflicts. Simply put, a write conflict is when many concurrently running threads need to write to the same memory address at the same time. If no processing is done, the previously written result may be overwritten by the latter, resulting in an incorre...

AI Technical Summary

Problems solved by technology

In addition, in the process of parallel computing, the thread load is extremely unbalanced, which seriously wastes the thread resources of the processor.

Images