FWIW the openib btl is mostly thread safe at this point. There is an
outstanding bug in the connection manager (udcm) that causes an abort in
debug builds but it should work in optimized builds. I have a patch in
https://github.com/open-mpi/ompi/pull/826 . You will need to specify
-mca btl_base_th
All,
It works!! Giles with the fix!
I ran it with his suggested flags:
mpirun --mca mtl ^psm -np 1 java MPITestBroke data/
The test code now runs without the segfault occurring around the 5th loop.
It will be a while before I can put this back into our bigger code that
first caused our segfault,
