hkaiser:  doesn't seem like a no-op. I'm getting the following runtime error:
Returned value from direct API:PID: 10993 on f3lix ready for attaching debugger. Once attached set i = 1 and continue
terminate called after throwing an instance of 'std::runtime_error'
  what():  cudaStreamSynchronize(m_stream) error( cudaErrorAssert): device-side assert triggered /home/nk/src/kokkos/core/src/Cuda/Kokkos_Cuda_Instance.cpp:314
Traceback functionality not available

terminate called recursively