Andrey Prokopenko
Andrey Prokopenko
I retested on Frontier with rocm 5.2 and 5.3, and can no longer observe any difference in DBSCAN and MST benchmark timings (between `master` b05b5f74 and this branch 65408d02).
One test in the CUDA-aware MPI build failed: ``` ArborX version: 1.7 (dev) ArborX hash : 65408d02 Kokkos version: 4.3.0 Running with arguments: perform knn search : true perform radius...
This has to be postponed until we figure out how ForTrilinos fits with Trilinos.
This PR won't be merged in the current form. Instead, the ideas in this PR will be incorporated into the refactoring for APIv2.
I redid the example based on the changes in ArborX in the recent months. It eliminated a lot of boiler plate code.
> Thoughts on whether and what to print to hints that something happened when one run the example. Probably should. Otherwise, it is a silent run.
Closing as we now are overhauling interface to allow this in a natural way through storing values.
Closing for inactivity. We also now have MLS implemented, though the distributed part has not been merged yet.
We should introduce `-Werror` to CUDA builds.
Should do it after kokkos/kokkos#7021 is resolved.