Parallel Peak Pruning for Scalable SMP Contour Tree Computation

As data sets grow to exascale, automated data analysis and visu- alisation are increasingly important, to intermediate human under- standing and to reduce demands on disk storage via in situ anal- ysis. Trends in architecture of high performance computing sys- tems necessitate analysis algorithms to make effective use of com- binations of massively multicore and distributed systems. One of the principal analytic tools is the contour tree, which analyses rela- tionships between contours to identify features of more than local importance. Unfortunately, the predominant algorithms for com- puting the contour tree are explicitly serial, and founded on serial metaphors, which has limited the scalability of this form of analy- sis. While there is some work on distributed contour tree computa- tion, and separately on hybrid GPU-CPU computation, there is no efficient algorithm with strong formal guarantees on performance allied with fast practical performance. We report the first shared SMP algorithm for fully parallel contour tree computation, with for- mal guarantees of O(lgnlgt) parallel steps and O(nlgn) work, and implementations with up to 10× parallel speed up in OpenMP and up to 50× speed up in NVIDIA Thrust.