https://hgpu.org/?p=7717
DMA-Assisted, Intranode Communication in GPU Accelerated Systems