Matrix transpose on meshes with buses

Abstract

In this paper we analyze the matrix transpose problem for 2- and 3-dimensional mesh architectures with row and column buses. First we consider the 2-dimensional problem, and we give a lower bound of approximately 0.45n for the number of steps required by any matrix transpose algorithm on an n×n mesh with buses. Next we present an algorithm which solves this problem in less than 0.5n+9 steps. Finally, we prove that the given lower bound remains valid for the 3-dimensional case as well.