Question

Using device_vector:

thrust::device_vector< int > iVec;
int* iArray = thrust::raw_pointer_cast( &iVec[0] );

but how can I do it if I have an array of device_vectors?

thrust::device_vector<int> iVec[10];

Ideally I would like to pass my array of device_vector to a 1D array to be handled on a CUDA kernel. Is it possible?

Answer 1

If I have understood your question correctly, what you are really try to do is create an array of raw pointers from an array of thrust::device_vectors. You should be able to do this like so:

const int N = 10;
thrust::device_vector<int> iVec[N];

int * iRaw[N];
for(int i=0; i<N; i++)
    iRaw[i] = thrust::raw_pointer_cast(iVec[i].data());

int ** _iRaw;
size_t sz = sizeof(int *) * N;
cudaMalloc((void ***)&_iRaw, sz);
cudaMemcpy(_iRaw, iRaw, sz, cudaMemcpyHostToDevice);

[disclaimer: written in browser, never compiled, never tested, use at own risk]

In the above code snippet, _iRaw holds the raw pointers of each of the device vectors in iVec. You could pass that to a kernel if you really wanted to.