Optimized upper bound for the scan of warpResults in the CUDA parallel scan

   // Perform the scan of warpResults using one warp.

   if( warpIdx == 0 )

      #pragma unroll

      for( int stride = 1; stride < Cuda::getWarpSize(); stride *= 2 ) {

      for( int stride = 1; stride < blockSize / Cuda::getWarpSize(); stride *= 2 ) {

         if( threadInWarpIdx >= stride )

            warpResults[ threadIdx.x ] = reduction( warpResults[ threadIdx.x ], warpResults[ threadIdx.x - stride ] );

         __syncwarp();