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

A parallel processing device for performing high-speed processing of tridiagonalization of a real symmetric matrix in a shared memory scalar parallel computer including a plurality of processors, said device comprising:

• a plurality of processors;
• an unit that performs block tridiagonalization process in which each of a plurality of column block matrices is tridiagonalized, comprising:
• a unit that divides a lower triangular matrix of a real symmetric submatrix that is utilized in the block tridiagonalization process in a column direction;
• a unit that allocates the divided lower triangular matrix of the real symmetric submatrix to the processors;
• a unit that makes each of the processors load a corresponding one of the divided lower triangular matrices in a single loading operation;
• and
• a unit that makes each of the processors perform computation in a row direction and the column direction of the corresponding one of the divided lower triangular matrices simultaneously in the computation of matrix-vector product in the block tridiagonalization process;
• and
• a unit that updates the lower triangular block submatrix by replacing the lower triangular block submatrix with a lower triangular block matrix of a second real symmetric submatrix that is obtained by removing a corresponding column block matrix that is obtained as the result of the block tridiagonalization process from the column block matrices.