#include <stdio.h>

__global__ void mykernel()
{
  // Declare shared variable with modifier __shared__.
  __shared__ int x;
  int tid = threadIdx.x;

  x = 17;

  // Synchronize all threads of this block.
  __syncthreads();

  // Thread 0 writes shared variable.
  if(tid == 0) x = 42;

  // Synchronize all threads of this block.
  __syncthreads();

  // Thread 1 reads shared variable.
  if(tid == 1) printf("thread 1 reads shared variable: %d\n",x);
}

int main()
{
  // Launch kernel with 1 block and 2 threads.
  mykernel<<<1, 2>>>();
  cudaDeviceSynchronize();
  return 0;
}