Reduction unrolling problem cuda Reduction unrolled

Accelerated Computing CUDA CUDA Programming and Performance
1

this is the original code I used for reduction -

	for (unsigned int s = (BLOCK_SIZE * BLOCK_SIZE) / 2; s>0; s>>=1)
	{
		if ((BLOCK_SIZE * ty + tx) < s)
			temps[BLOCK_SIZE * ty + tx] += temps[BLOCK_SIZE * ty + tx+s];
		__syncthreads();
	}
	float VarienceS = temps[0];

this is the unrolled code by cuda, and it gives me differnt outputs the the original , and I dont know why -

  if (BLOCK_SIZE * ty + tx < 128)   temps[BLOCK_SIZE * ty + tx]+= temps[BLOCK_SIZE * ty + tx + 128]; 
   __syncthreads(); 
  if (BLOCK_SIZE * ty + tx < 64)  temps[BLOCK_SIZE * ty + tx]+= temps[BLOCK_SIZE * ty + tx + 64]; 
  __syncthreads();

   if (BLOCK_SIZE * ty + tx < 32)
   {
		  temps[BLOCK_SIZE * ty + tx] +=temps[BLOCK_SIZE * ty + tx + 32];	  
		  temps[BLOCK_SIZE * ty + tx] += temps[BLOCK_SIZE * ty + tx + 16];		 		 
		  temps[BLOCK_SIZE * ty + tx] += temps[BLOCK_SIZE * ty + tx +  8];	  		  
		  temps[BLOCK_SIZE * ty + tx] += temps[BLOCK_SIZE * ty + tx +  4];			  
		  temps[BLOCK_SIZE * ty + tx] += temps[BLOCK_SIZE * ty + tx +  2]; 			  
		  temps[BLOCK_SIZE * ty + tx] += temps[BLOCK_SIZE * ty + tx +  1]; 
	  
   }

Num Of Threads is known and is 256 ,
BLOCKSIZE IS 16