cuda faster to initialise in host or in device code example

Example 1: cuda copy memory

cudaMemcpy(d_dst, h_src, N*sizeof(float), cudaMemcpyHostToDevice);
cudaMemcpy(h_dst, d_src, N*sizeof(float), cudaMemcpyDeviceToHost);

Example 2: cuda allocate memory

const size_t SIZE = 900;
  float *abc;
  cudaMalloc((void **)&abc, SIZE * SIZE * sizeof(float));