GPUkernelfunction<<<blocknum, threadnum>>>()

nvcc -arch=sm_70 -o out some-CUDA.cu -run  

#include <stdio.h>

__global__ void loop()
{
    printf("This is iteration number %d %d\n", blockIdx.x, threadIdx.x);
}

int main()
{
  /* Loop over 2 blocks with 10 threads each */

  loop<<<2, 10>>>();
  cudaDeviceSynchronize();
}

// CPU-only

int N = 2<<20;
size_t size = N * sizeof(int);

int *a;
a = (int *)malloc(size);

// Use `a` in CPU-only program.

free(a);
// Accelerated

int N = 2<<20;
size_t size = N * sizeof(int);

int *a;
// Note the address of `a` is passed as first argument.
cudaMallocManaged(&a, size);

// Use `a` on the CPU and/or on any GPU in the accelerated system.

cudaFree(a);

__global__ void kernel(int *a, int N)
{
  int indexWithinTheGrid = threadIdx.x + blockIdx.x * blockDim.x;
  int gridStride = gridDim.x * blockDim.x;

  for (int i = indexWithinTheGrid; i < N; i += gridStride)
  {
    // do work on a[i];
  }
}