Cyclic reduction on distributed shared memory machines
S. Allmann, Thomas Rauber, Gudula Rünger · 2002
Cyclic reduction for the solution of linear equation systems with banded matrices exhibits fine to medium grain potential parallelism with regular but diverse data dependencies. We consider the parallel implementation for this algorithm on a distributed shared memory machine with different programming models. As distributed shared memory machine we use the Convex SPP2000. We compare the runtime results with results from a Cray T3E.