Function prefetch_async 

pub fn prefetch_async(
    dev_ptr: CUdeviceptr,
    bytes: usize,
    target: PrefetchTarget,
    stream: &Arc<CudaStream>,
) -> Result<(), GpuError>

Prefetch the byte range [dev_ptr .. dev_ptr+bytes) to target, issued onto stream. Wraps cuMemPrefetchAsync_v2.

Returns Unrecoverable on hosts where libcuda.so isn’t loadable.