一些人已经举了一些例子，其中涉及到创建一个新数组。

还有一些需要考虑的事情:

(a)不实际移动数据，只需以不同的方式遍历“旋转”的数组。

(b)就地轮换可能有点棘手。您需要一点空白的地方(大概相当于一行或一列的大小)。有一篇古老的ACM论文是关于进行原地转置的(http://doi.acm.org/10.1145/355719.355729)，但是他们的示例代码是令人讨厌的充满goto的FORTRAN。

附录:

http://doi.acm.org/10.1145/355611.355612是另一种更优越的就地转置算法。

试试我图书馆的算盘——常见的:

@Test
public void test_42519() throws Exception {
    final IntMatrix matrix = IntMatrix.range(0, 16).reshape(4);

    N.println("======= original =======================");
    matrix.println();
    // print out:
    //    [0, 1, 2, 3]
    //    [4, 5, 6, 7]
    //    [8, 9, 10, 11]
    //    [12, 13, 14, 15]

    N.println("======= rotate 90 ======================");
    matrix.rotate90().println();
    // print out:
    //    [12, 8, 4, 0]
    //    [13, 9, 5, 1]
    //    [14, 10, 6, 2]
    //    [15, 11, 7, 3]

    N.println("======= rotate 180 =====================");
    matrix.rotate180().println();
    // print out:
    //    [15, 14, 13, 12]
    //    [11, 10, 9, 8]
    //    [7, 6, 5, 4]
    //    [3, 2, 1, 0]

    N.println("======= rotate 270 ======================");
    matrix.rotate270().println();
    // print out:
    //    [3, 7, 11, 15]
    //    [2, 6, 10, 14]
    //    [1, 5, 9, 13]
    //    [0, 4, 8, 12]

    N.println("======= transpose =======================");
    matrix.transpose().println();
    // print out:
    //    [0, 4, 8, 12]
    //    [1, 5, 9, 13]
    //    [2, 6, 10, 14]
    //    [3, 7, 11, 15]

    final IntMatrix bigMatrix = IntMatrix.range(0, 10000_0000).reshape(10000);

    // It take about 2 seconds to rotate 10000 X 10000 matrix.
    Profiler.run(1, 2, 3, "sequential", () -> bigMatrix.rotate90()).printResult();

    // Want faster? Go parallel. 1 second to rotate 10000 X 10000 matrix.
    final int[][] a = bigMatrix.array();
    final int[][] c = new int[a[0].length][a.length];
    final int n = a.length;
    final int threadNum = 4;

    Profiler.run(1, 2, 3, "parallel", () -> {
        IntStream.range(0, n).parallel(threadNum).forEach(i -> {
            for (int j = 0; j < n; j++) {
                c[i][j] = a[n - j - 1][i];
            }
        });
    }).printResult();
}

我只用一个循环就能做到。时间复杂度看起来像O(K)其中K是数组中的所有元素。 下面是我用JavaScript做的:

首先，我们用一个数组来表示n^2矩阵。然后，像这样迭代它:

/**
 * Rotates matrix 90 degrees clockwise
 * @param arr: the source array
 * @param n: the array side (array is square n^2)
 */
function rotate (arr, n) {
  var rotated = [], indexes = []

  for (var i = 0; i < arr.length; i++) {
    if (i < n)
      indexes[i] = i * n + (n - 1)
    else
      indexes[i] = indexes[i - n] - 1

    rotated[indexes[i]] = arr[i]
  }
  return rotated
}

基本上，我们转换源数组下标:

[0,1,2,3,4,5,6,7,8] => [2,5,8,1,4,7,0,3 6]

然后，使用这个转换后的索引数组，我们将实际值放在最终旋转的数组中。

下面是一些测试用例:

//n=3
rotate([
  1, 2, 3,
  4, 5, 6,
  7, 8, 9], 3))

//result:
[7, 4, 1,
 8, 5, 2,
 9, 6, 3]


//n=4
rotate([
  1,  2,  3,  4,
  5,  6,  7,  8,
  9,  10, 11, 12,
  13, 14, 15, 16], 4))

//result:
[13,  9,  5,  1,
 14, 10,  6,  2,
 15, 11,  7,  3,
 16, 12,  8,  4]


//n=5
rotate([
  1,  2,  3,  4,  5,
  6,  7,  8,  9,  10,
  11, 12, 13, 14, 15,
  16, 17, 18, 19, 20,
  21, 22, 23, 24, 25], 5))

//result:
[21, 16, 11,  6,  1, 
 22, 17, 12,  7,  2, 
 23, 18, 13,  8,  3, 
 24, 19, 14,  9,  4, 
 25, 20, 15, 10,  5]

这是c#的

int[,] array = new int[4,4] {
    { 1,2,3,4 },
    { 5,6,7,8 },
    { 9,0,1,2 },
    { 3,4,5,6 }
};

int[,] rotated = RotateMatrix(array, 4);

static int[,] RotateMatrix(int[,] matrix, int n) {
    int[,] ret = new int[n, n];

    for (int i = 0; i < n; ++i) {
        for (int j = 0; j < n; ++j) {
            ret[i, j] = matrix[n - j - 1, i];
        }
    }

    return ret;
}

虽然旋转数据可能是必要的(也许是为了更新物理存储的表示)，但在数组访问上添加一层间接层(也许是一个接口)会变得更简单，可能更性能:

interface IReadableMatrix
{
    int GetValue(int x, int y);
}

如果你的矩阵已经实现了这个接口，那么它可以通过这样一个装饰器类来旋转:

class RotatedMatrix : IReadableMatrix
{
    private readonly IReadableMatrix _baseMatrix;

    public RotatedMatrix(IReadableMatrix baseMatrix)
    {
        _baseMatrix = baseMatrix;
    }

    int GetValue(int x, int y)
    {
        // transpose x and y dimensions
        return _baseMatrix(y, x);
    }
}

旋转+90/-90/180度，水平/垂直翻转和缩放都可以以这种方式实现。

Performance would need to be measured in your specific scenario. However the O(n^2) operation has now been replaced with an O(1) call. It's a virtual method call which is slower than direct array access, so it depends upon how frequently the rotated array is used after rotation. If it's used once, then this approach would definitely win. If it's rotated then used in a long-running system for days, then in-place rotation might perform better. It also depends whether you can accept the up-front cost.

与所有性能问题一样，测量，测量，测量!

下面是Java语言:

public static void rotateInPlace(int[][] m) {
    for(int layer = 0; layer < m.length/2; layer++){
        int first = layer;
        int last = m.length - 1 - first;
        for(int i = first; i < last; i ++){
            int offset = i - first;
            int top = m[first][i];
            m[first][i] = m[last - offset][first];
            m[last - offset][first] = m[last][last - offset];
            m[last][last - offset] = m[i][last];
            m[i][last] = top;
        }
    }
}