How to allocate and copy a complex object on device memory?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi everyone,

I am pretty new to CUDA. I would like to have some guidance if you don’t mind please. In fact, i would like to allocate memory for a complex object on device memory and then copy this object from host to device. My struct looks something like :

struct dummy{
   int size;
   unsigned int *data;
   
   dummy(int s) 
           : size(s), data(new unsigned int[s])
	{}
   void fill(){
      //fill the data array with random values
   }
}

On the CPU I have created and filled my dummy object :

dummy obj_cpu(80);
obj_cpu.fill();

Now I would like to allocate sufficient memory on the device for obj_cpu and copy obj_cpu to it. Thus I thought of doing this :

// Declare a dummy pointer on the CPU
dummy obj_gpu;
// Allocate memory for obj_gpu on device memory
cudaMalloc(&obj_gpu, sizeof(obj_cpu));
// Copy	data from host(CPU) to device(GPU) as a bulk
cudaMemcpy(&obj_gpu, &obj_cpu, sizeof(obj_gpu), cudaMemcpyHostToDevice);

Is this the right way to allocate memory/copy complex object on the device memory or should I do it this way instead :

// Declare a dummy pointer on the CPU
dummy obj_gpu;
// Allocate memory for obj_gpu.data on device memory
cudaMalloc(&(obj_gpu.data), obj_cpu.s*sizeof(unsigned int));
// Copy	data from host(CPU) to device(GPU) element by element
obj_gpu.s=obj_cpu.s;
for(int j=0; j<obj_cpu.s;j++)
        cudaMemcpy(&(obj_gpu.data[j]), &(obj_cpu.data[j]), sizeof(unsigned int), cudaMemcpyHostToDevice);

I would love to have please some feed-backs on which strategy should I use and also why. Thank you

2

Check here:
https://stackoverflow.com/questions/14284964/cuda-how-to-allocate-memory-for-data-member-of-a-class

3

@saulocpp : Thank you