CUDA：为什么将共享内存阵列填充一列会使内核速度降低250％？

Question

 __global__ void setRowReadColPad(int *out)
 {
    // static shared memory
    __shared__ int tile[32][33];

    // mapping from thread index to global memory offset
    unsigned int idx = threadIdx.y * blockDim.x + threadIdx.x;


    // shared memory store operation
    tile[threadIdx.y][threadIdx.x] = idx;


    // wait for all threads to complete
    __syncthreads();

    // shared memory load operation

    out[idx] = tile[threadIdx.x][threadIdx.y];

 }

使用共享内存填充需要13.473us

不进行填充需要5.025us

有人可以解释一下为什么填充对内核运行时有如此大的影响吗？