#include <cuda.h>
#include <stdio.h>

#define N 10

__global__ void kernel(int *result) { return; }

int main() {
  int **result;
  cudaMallocManaged(&result, N * sizeof(int*));
  for(int i = 0; i < N; i++) cudaMallocManaged(&result[i], sizeof(int));

  cudaDeviceSynchronize();
  printf("okay\n");

  for(int i = 0; i < N; i++) {
    kernel<<<1, 1>>>(result[i]);
    // cudaDeviceSynchronize();  // decomment me, and evertying works
  }

  cudaDeviceSynchronize();
  printf("perfect\n");

  return 0;
}