Closed dschwen closed 6 years ago
There are some iffy bits in the design of KDTree. It is not guaranteed that the index list is fully populated (it is padded with max uint). In SlaveNeighborhoodThread the entire list is used to index into nodes. This can potentially segfault. I'll fix this when I switch to the new api stuff.
Rationale
Remove unnecessary copying, container creation on the stack, and use new nanoflann API features (libmesh/libmesh#1563).
Description
We get the number of matched points (no need to screw around with limits max). We can pass the pointer to the internal data of
Point
rather than copy it into a C array. We should use Utility::pow rather than std::pow. We should have a permanent container for the distance dummy data. I'd also like to add an API to get the distance data, but maybe not in this issue.Impact
Performance improvements.