OpenCL and DMA Data Transfer

Question

The AMD APP Programming Guide says that (p.no 4-15):

For transfers <= 32 kB: for transferring from the host to the device, the data is copied by the CPU to the memory buffer with a fixed host, and the DMA mechanism transfers data to the device’s memory. The opposite is done for transferring from a device to a host.

Is the aforementioned DMA, DMA CPU processor, or DMA GPU engine?

Answer 1

I believe that this is a DMA GPU processor, as on some cards (for example, NVIDIA) you can read and write at the same time (so this is a GPU function, not a processor capability).