Use GPUDirect to DMA from GPU to PCIe device

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I am looking at (GPUDirect RDMA :: CUDA Toolkit Documentation), however I have a few questions that don’t seem to be described (or perhaps I’m just missing it). The page mostly describes how to set up a PCIe device driver to request DMA transfers:

  1. How would one program the GPU itself to forward completed frames to the PCIe device BAR?
  2. How is a PCIe device informed of the start / end of a frame for synchronization purposes?
  3. Ideally I would like to use this with the Tesla P6. Is there a way to know what the BAR addresses of the virtual framebuffers are?

Thanks in advance. I really just want to be able to read frames from the GPU over PCIe. I hope my questions don’t sound too silly.