Build (or) Send a Tree on/to the GPU

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I am currently looking at tree and have the following structure

typedef struct nd{

    short int final;
    char number; 
    struct nd *child[2]; 

}node;

So my plan was to build the tree on the GPU, I get how to allocate memory, since I know exactly how many nodes I will have in there and can therefor avoid synchronisation, although I don’t see how to build the tree on the GPU.

Can I simply make some code like :

__global__ void Tree((node * tree, int size)) {

--> That's the part where I'm actually stuck.  How can I build my tree here 

}

int main(int argc, char **argv){

[...]

        cudaMalloc( (void**)&treeCUDA, sizeof(node)*256*256 );

	Tree<<<1,256>>>(treeCUDA, size);


[...]

}

Thanks for the help, I have also read that a tree could be contained in an array, are there any examples anywhere on how to do that ?

Thanks.