/*WBL 14 March 2015 taken from
http://nvlabs.github.io/nvbio/host_device_page.html
*/
/*compile with? 

Still no joy try stealing command line from qmap

/usr/local/cuda/bin/nvcc host_device_page.cu -ccbin /opt/centos/devtoolset-1.1/root/usr/bin/cc -m64 -DPLATFORM_X86 -Xcompiler ,\"-Wall\",\"-Wno-unknown-pragmas\",\"-Wstrict-aliasing=0\",\"-fopenmp\",\"-O3\",\"-DNDEBUG\",\"-msse4.2\",\"-mpopcnt\",\"-funroll-loops\" --gpu-architecture=sm_35 -use_fast_math -DNVCC -I/usr/local/cuda/include -I../nvbio-master -I../nvbio-master/contrib -I../nvbio-master/build/contrib/zlib-1.2.7 -I../nvbio-master/contrib/crc -I../nvbio-master/contrib/bamtools -I../nvbio-master/contrib/moderngpu/include -I/usr/local/cuda/include


*/


//taken from hello_world.cu
//

#include <stdio.h>
#include <stdlib.h>
#include <nvbio/basic/packed_vector.h>
#include <nvbio/basic/dna.h>
#include <nvbio/strings/infix.h>
#include <nvbio/strings/seeds.h>


using namespace nvbio;


//    however, this won't be possible in CUDA until UVM is finally available. With NVBIO, you'd do this instead:


    __global__ void my_kernel( // the CUDA kernel
    nvbio::vector_view<uint32> vec) // NVBIO's surrogate of a reference
    {
    const uint32 tid = threadIdx.x + blockIdx.x * blockDim.x; // compute a linear thread id
    if (tid < vec.size())
    vec[tid] = tid * 10;
    }
    int main()
    {
    nvbio::vector<device_tag,uint32> vec( 1000000 );
    const uint32 blockdim = 128;
    const uint32 n_blocks = util::divide_ri( vec.size(), blockdim );
    my_kernel<<<n_blocks,blockdim>>>( nvbio::plain_view( vec ) );
    }

//  This basic pattern can be applied to all of NVBIO's data structures that are meant to be setup from the host and accessed from the device.