问为什么CUDA同步点不能阻止比赛条件？

EN

========= Race reported between Read access at 0x00004098 CUDA.cu:123:KernelFunction()
=========     and Write access at 0x00005058 in CUDA.cu:146:KernelFunction() [529996 hazards]  

// Synchronization Point 1 
__syncthreads(); 
__threadfence_block();

   for (i = 0; i < COUNT0; i++) {
       // Synchronization Point 0
       __syncthreads();
       __threadfence_block();
       for (j = 0; j < COUNT1; j++) {
          index = j*blockDim.x + threadIdx.x;
          if (index < THREAD_COUNT0) {
             for (k = 0; k < COUNT2; k++)
                sharedMemoryIndex = function0(index);
                value = sharedMemory0[sharedMemoryIndex];
             }
          }         
       }

       // Synchronization Point 1
       __syncthreads();
       __threadfence_block();
       for (j = 0; j < COUNT2; j++) {
          index = j*blockDim.x + threadIdx.x;
          if (index < THREAD_COUNT1) {
            sharedIndex0 = function1(index);
            sharedIndex1 = function2(index);
            sharedMemory0[sharedIndex0] = sharedMemory1[sharedIndex1];
          }
       }
    }

========= Barrier error detected. Encountered barrier with divergent threads in block
=========     at 0x00004ad8 in CUDA.cu:139:KernelFunction()
=========     by thread (0,0,0) in block (8,0,0)