Enabling Fast, Noncontiguous GPU Data Movement in Hybrid MPI+GPU Environments | IEEE Conference Publication | IEEE Xplore