Andrey Prokopenko
Andrey Prokopenko
Force pushed to remove the custom layouts specialization.
Let's merge this, and I'll push compile time stride in the follow up PR.
Other than CTAD, everything else has been implemented.
We still have some inconvenient and confusing stuff. If a user is really only interested in indices (aka, APIv1), we have them hopping through: - Having to call `attach_indices` -...
I need to check that FDBSCAN is not affected. We don't have benchmark for neighbor lists, so can't say how much it affected it. But I expect that there is...
``` saturn dbscan $ ./ArborX_master_7abc0268 --binary --filename ~/data/hacc_497M.arborx --eps 0.014 --core-min-size 2 --impl fdbscan --verbose -- construction : 0.371 -- query+cluster : 3.143 -- postprocess : 0.516 total time :...
Move a part of the PR (the one that isn't related to switching HalfTraversal API to v2) to #1024.
Rebased on #1024.
The following change seems to have fixed the performance issue: ```diff --- a/src/details/ArborX_DetailsHalfTraversal.hpp +++ b/src/details/ArborX_DetailsHalfTraversal.hpp @@ -52,7 +52,7 @@ struct HalfTraversal KOKKOS_FUNCTION void operator()(int i) const { - auto const...
@dalg24 ready for re-review