Isomorphic New Parallel Division Methods and Parallel Algorithms for Giant Matrix Transpose

Abstract

In this paper, the weakness of the traditional matrix division methods for giant matrix transpose is pointed; Specific to the nature of the giant matrixes’ transpose and based on the characteristics of asynchronous PRAM parallel computing model, some new division (such as booklet belt division, closed-loop division and Checkerboard-belt-shaped compound division) methods are proposed for improving the traditional existing matrix division methods, and their new parallel transpose algorithms for giant matrixes are advanced; The correspondence pressure between machines computational capability and enormously computing quantity in the parallel transpose process are reduced, while the parallel processing operating efficiency are enhanced, and the cost of parallel processing realization should be brought down by these new parallel algorithms (which could be realized on the cluster of workstations) based on division methods used fully the symmetrical characteristic of the giant matrix transpose operation.