Parallel processing method of tridiagonalization shared memory scalar parallel computer for, of a real symmetric matrix

摘要

When performing a shared memory scalar parallel computer tridiagonalization of real symmetric matrix, is performed by dividing each thread updates the matrix. At this time, since the symmetric matrix is ​​a matrix, is calculated by using the data of the lower triangular matrix portion. The lower triangular matrix components can be longitudinally divided, assigned to each CPU, but this time, to assign the number of elements to be assigned to each CPU to be the same. Then, a load of one-time data, to perform operations simultaneously and transverse vertical operation, to reduce the number of memory load. Lower triangular matrix portion of the diagonal block matrix portion is an update while recursively be divided into smaller lower triangular matrix and the two small square matrix one.