This function is the thread-specific version of the shmem_fence
routine. The function ensures ordering of put (remote write) and
non-blocking get operations. All such operations, which are issued by
the calling thread to a particular remote processing element (PE) prior
to the call to shmemx_thread_fence, are guaranteed to be delivered
before any subsequent such operations to the same PE that follow the
call to shmemx_thread_fence and are issued by the calling thread.
If this routine encounters an error, it will print an error message and
abort.