作者: J Tierny , J Jomier , P. Fortin , Charles Gueunet
DOI:
关键词:
摘要: This paper presents a new algorithm for the fast, shared memory, multi-core computation of augmented contour trees on triangulations. In contrast to most existing parallel algorithms our technique computes trees, enabling full extent tree based applications including data segmentation. Our approach completely revisits traditional, sequential re-formulate all steps as set independent local tasks. includes procedure Fibonacci heaps join and split two intermediate structures used compute tree, whose constructions are efficiently carried out concurrently thanks dynamic scheduling task parallelism. We also introduce combination these into output global tree. Overall, this results in superior time performance practice, both OpenMP runtime. report numbers that compare reference multi-threaded implementations merge trees. These experiments demonstrate run-time efficiency its scalability common workstations. utility segmentation applications.