Segfault in Distributor
Created by: aprokop
#222 (checked through bisection) seems to have introduced a new bug, that results in segfault when running distributed benchmark with CUDA backend. I am able to reproduce it even running as
mpirun -np 2 ./ArborX_DistributedTree.exe --node cuda --values=1 --queries=1
As long as at least one search is performed (does not matter radius or nearest), it segfaults. The backtrace that I've seen points to doPostsAndWaits
in Distributor, specifically to MPI_Waitall
call.
@masterleinad I hope you can help me with this.