透過您的圖書館登入
IP:3.137.187.233

並列摘要


The mesh is an architecture that has many scientific applications, and matrix transpose is an important permutation frequently performed in various techniques involving systems of linear equations. In this paper, we present an optimal algorithm for performing matrix transpose on meshes that support wormhole switching. If N is even, our algorithm takes (The equation is abbreviated) communication steps to perform matrix transpose on an N×N mesh and requires only 3 more steps when the routing is restricted to XY routing, which is supported by most commercial mesh-connected parallel computers. The lower bound is (The equation is abbreviated) and the best previous bound is about N/3.27. The complexity of our algorithm almost matches the lower bound. Furthermore, our algorithm is simple and can be implemented on current mesh-connected parallel computers.

延伸閱讀