Many-core architecture-oriented sparse matrix vector multiplication many-core optimization method

Abstract

The invention discloses a many-core architecture-oriented sparse matrix vector multiplication many-core optimization method. The method comprises the following steps of S1, knowing a sparse matrix A of which the row number is m and the column number is n and a vector x of which the length is n; solving a vector y with the length of m, wherein y = Ax is the dot product of the sparse matrix A and the vector x; the method comprises the following steps: S1, defining the size blkxsize of an x vector block, and partitioning the x vector element according to an x vector element subscript to partitionan x vector; and S2, counting the number of the x vector block corresponding to the column number of each row of non-zero elements in the original sparse matrix, namely the sparse matrix A. Accordingto the blocking information of the x vector, namely the number information of the x vector block where the x vector element obtained by solving in the step S1 is located, thereby counting the numberinformation of the x vector block required by each row of the sparse matrix when the sparse matrix vector is multiplied. According to the method, the overall many-core acceleration performance is improved, the locality of data access is improved, and the optimization effect on unstructured grid CFD application is obvious.

Description

technical field [0001] The invention belongs to the technical field of sparse matrix-vector multiplication, and in particular relates to a multi-core optimization method for sparse matrix-vector multiplication oriented to many-core architecture. Background technique [0002] With the deepening of unstructured grid CFD application research and the rapid development of supercomputer technology, the many-core acceleration of sparse matrix-vector multiplication has become one of the focuses of CFD application optimization research. [0003] Due to the loose distribution of non-zero elements of the sparse matrix generated by the CFD application of unstructured grids, and the large span of different non-zero element numbers, one of the calculation cores of the program - Sparse Matrix-Vector Multiplication (Spmv) has a very obvious discrete memory access problem. This has also become a difficulty in the many-core optimization of non-institutional grid CFD applications. [0004] As...

AI Technical Summary

Problems solved by technology

This method can effectively alleviate the discrete memory access problem in sparse matrix-vector multiplication. However, the sparse matrices generated in unstructured grid CFD applications are often extremely sparse (each row of a sparse matrix often has only a few non-zero elements) and the number of non-zero elements The distribution has a certain statistical law

Due to the extremely sparse matrix, the traditional fixed-width row-column block method will have very serious load balancing problems, and cannot effectively utilize the data transmission bandwidth between the CPU and the many-core coprocessor, resulting in the Spmv many-core optimization for CFD applications. not effectively

Images