Mapping FPGA Virtex5/7 memory to the UVA for data transfer GPU-FPGA via DMA and GPU-Cores

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello.

Are there any drivers for Linux-x86_64 and Verilog for Virtex5/7 for mapping FPGA memory to the physical memory addressing of CPU-x86_64 and to the virtual addressing of current process via PCIe gen2/3 8x, and then mapping it to the CUDA UVA for using GPU-DMA for GPU<->FPGA data transfer and access from GPU-Cores to the FPGA memory?

And then copy from/to this memory area from CPU by using:

  • x86 instructions: MOV, REP MOVS
  • x86_64_SSE: MOVNTDQ, MOVNTDQA
  • GPU-DMA for direct access from/to GPU-RAM to the FPGA-RAM by using
    cudaMemcpy(); – ( CUDA UVA (Unified Virtual Addressing) )

There is what I need, but is there somewhere like this under Linux?
Virtex5 + Win Driver: in PDF-file - Fig. 5 GPU to FPGA DMA code sample;:

Or what of these can I use?

2

Any follow-up for this question？