#include <cuda_runtime.h>
#include <iostream>

__global__ void kernel() {
  __shared__ volatile bool looped;
  __shared__ volatile bool done;

  if (!threadIdx.x) {
    looped = false;
    done   = false;
  }

  while (true) {
    if ((threadIdx.x == 0) && looped)
      done = true;

    if (done)
      return;

    if (threadIdx.x == 0)
      looped = true;
  }
}

int main() {
  kernel<<<dim3(1), dim3(2)>>>();
}