Reintroduce sync, and set the total on every thread.

Author: andrew-hardin

src/popsift/common/excl_blk_prefix_sum.h

@@ -137,12 +137,10 @@ class Block
             if( threadIdx.y == 0 && threadIdx.x == 31 ) {
                 loop_total += ibs;
             }
+            __syncthreads();
         }
 
-        if( threadIdx.y == 0 && threadIdx.x == 31 )
-        {
-            _total_writer.set( loop_total );
-        }
+        _total_writer.set( loop_total );
     }
 };