CUDA kernel使用printf调试，一定要加cudaDeviceSynchronize

#include <stdio.h>

__global__ void print() {
  printf("Hello from (%d, %d)\n", blockIdx.x, threadIdx.x);
}

void test() {
  print<<<4, 4>>>();
  cudaDeviceSynchronize();
  return;
}

int main() {
  test();
  return 0;
}