I’m trying to use Mapped-memory and the so called zero-copy, but I’m having a few problems with it…
I’m trying to do this:
cudaHostAlloc((void**)&stream_in, sizeof(soa), cudaHostAllocMapped); checkCUDAError("cudaHostAllocMapped");
Which results in a: “Cuda error: cudaHostAllocMapped: feature is not yet implemented.”.
I’m using the 2.2SDK and Toolkit so this should work, right?