这是一个空间旋转方法，由java编写，只适用于正方形。对于非正方形的2d数组，无论如何都必须创建新数组。

private void rotateInSpace(int[][] arr) {
    int z = arr.length;
    for (int i = 0; i < z / 2; i++) {
        for (int j = 0; j < (z / 2 + z % 2); j++) {
            int x = i, y = j;
            int temp = arr[x][y];
            for (int k = 0; k < 4; k++) {
                int temptemp = arr[y][z - x - 1];
                arr[y][z - x - 1] = temp;
                temp = temptemp;

                int tempX = y;
                y = z - x - 1;
                x = tempX;
            }
        }
    }
}

通过创建新数组旋转任何大小的2d数组的代码:

private int[][] rotate(int[][] arr) {
    int width = arr[0].length;
    int depth = arr.length;
    int[][] re = new int[width][depth];
    for (int i = 0; i < depth; i++) {
        for (int j = 0; j < width; j++) {
            re[j][depth - i - 1] = arr[i][j];
        }
    }
    return re;
}

原地旋转不可能比O(n²)更快，原因是如果我们想旋转矩阵，我们必须至少一次触及所有n²元素，无论你实现什么算法。

short normal[4][4] = {{8,4,7,5},{3,4,5,7},{9,5,5,6},{3,3,3,3}};

short rotated[4][4];

for (int r = 0; r < 4; ++r)
{
  for (int c = 0; c < 4; ++c)
  {
    rotated[r][c] = normal[c][3-r];
  }
}

简单的c++方法，尽管在大数组中会有很大的内存开销。

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!

#转置是Ruby的Array类的标准方法，因此:

% irb
irb(main):001:0> m = [[1, 2, 3, 4], [5, 6, 7, 8], [9, 0, 1, 2], [3, 4, 5, 6]]
=> [[1, 2, 3, 4], [5, 6, 7, 8], [9, 0, 1, 2], [3, 4, 5, 6]] 
irb(main):002:0> m.reverse.transpose
=> [[3, 9, 5, 1], [4, 0, 6, 2], [5, 1, 7, 3], [6, 2, 8, 4]]

实现是一个用c写的n^2转置函数，你可以在这里看到: http://www.ruby-doc.org/core-1.9.3/Array.html#method-i-transpose 通过选择“点击切换源”旁边的“转置”。

我记得比O(n^2)的解更好，但只适用于特殊构造的矩阵(如稀疏矩阵)

哦，伙计。我一直认为这是一个“我很无聊，我能思考什么”的谜题。我想出了我的原地换位码，但到了这里发现你的和我的几乎一模一样……啊,好。这里是Ruby版本。

require 'pp'
n = 10
a = []
n.times { a << (1..n).to_a }

pp a

0.upto(n/2-1) do |i|
  i.upto(n-i-2) do |j|
    tmp             = a[i][j]
    a[i][j]         = a[n-j-1][i]
    a[n-j-1][i]     = a[n-i-1][n-j-1]
    a[n-i-1][n-j-1] = a[j][n-i-1]
    a[j][n-i-1]     = tmp
  end
end

pp a