2D array inside a structure

Hello Community.

I was searching a solution to the following problem and I found something in the forum but still I cannot make it work (The Official NVIDIA Forums | NVIDIA ).

So the idea is to have an object for each execution unit that does all the calculations and store the results.

The objects may be also different each other.

The fact is that I don’t allocate correctly the arrays of my structure but from the other posts in the forum I don’t get how should I do that.

Could somebody go trough my code here and tell me how to fix it, please?

#include <stdlib.h>

#include <iostream>

#include <cuda.h>

#include <curand_kernel.h>

using namespace std;

struct test{

		float *d1_vals;   //an 1D array

		float **d2_vals;  //an 2D array

		int rows;

		int columns;

		void init(int x, int y); //init assign x to rows, y to columns, allocate d2_vals and d1_vals on the HOST

		__device__ void foo();   // device function to fill the arrays

		void print_r();            //a print function


//allocate the arrays of the structure

void test::init(int x, int y)




d1_vals=new float[rows];

d2_vals=new float*[rows];

for(int i=0;i<rows;i++)

  d2_vals[i]=new float[columns];


//print the arrays

void test::print_r()


	for(int j=0;j<rows;++j)



	  for(int k=0;j<columns;++k)




//foo simply fill d2_vals with numbers position_x*position_y and d1_vals with position_x, it have to be called after init

__device__ void test::foo()


	for(int j=0;j<rows;++j)



	  for(int k=0;j<columns;++k)




__global__ void fillvals(struct test *a)


const int id=blockIdx.x;



int main()


int num_obj=2, num_rows=2, num_col=3;

struct test *somestruct = (struct test*)malloc(num_obj*sizeof(struct test));

for(int h=0;h<num_obj;h++)


struct test *to_device;

cudaMalloc ( (void **)&to_device, num_obj*sizeof(struct test));

cout<<"before the kernel"<<endl;

for(int k=0;k<num_obj;k++)


cudaMemcpy(to_device,somestruct,num_obj*sizeof(struct test),cudaMemcpyHostToDevice);


cudaMemcpy(somestruct,to_device,num_obj*sizeof(struct test),cudaMemcpyDeviceToHost);

cout<<"after the kernel"<<endl;

for(int k=0;k<num_obj;k++)




This is a kind of “BUMP”.
Nobody answer to this thread so I am wondering if what I am asking is possible.
If somebody knows something it would be nice to know.

You are copying the structs to the device even though they contains host pointers. You also miss to copy the data pointed to.

The best option is to flatten everything into one big one-dimensional array, and use indices instead of pointers into it (you’ll need to add the sizes of all arrays before you can allocate the memory for it). Then you can copy the memory for the array in one go, and indices are the same in host and device memory, so you don’t need to care about fixing up pointers after copying the structs.