Combining Smoother and Residual Calculation in v-cycle AMG for Symmetric Problems
We examine a modified implementation of the v-cycle multigrid scheme which takes into account the memory traffic, i.e. the movement of data between the memory and the processing units. It is known that the relatively slow data transfer is responsible for the poor parallel performance of multigrid algorithms on certain shared memory architectures e.g. those with a front-side bus memory controller. The modification is simple but it speeds up computations by up to 15%.
KeywordsShared Memory Memory Bandwidth Total Computing Time Multigrid Algorithm Memory Access Pattern
Unable to display preview. Download preview PDF.
- 1.Emans, M., van der Meer, A.: Mixed-precision AMG as linear equation solver for definite systems. In: Sloot, P.M.A., Dongarra, G.V.A., Dongarra, J. (eds.) ICCS 2010, Part I. Procedia Computer Science, vol. 1, pp. 175–183. Elsevier Science, North Holland (2010)Google Scholar
- 5.Gee, M., Siefert, C., Hu, J., Tuminaro, R., Sala, M.: ML 5.0 Smoothed Aggregation User’s Guide. SAND 2006-2009 Unlimited Release (2006)Google Scholar
- 6.McCalpin, J.: Memory bandwidth and machine balance in current high performance computers. IEEE Computer Society Technical Committee on Computer Architecture (TCCA) Newsletter, 19–25 (December 1995)Google Scholar