java - Java中数组寻址的时间复杂度差异

Question

所以我在编码涉及时间复杂度的图像处理功能时有一个随机问题。以下是我的原始代码片段:

    long start = System.currentTimeMillis();

    for (int i = 0; i < newWidth; i++) {
        for (int j = 0; j < newHeight; j++) {

            double x = i * scaleX;
            double y = j * scaleY;

            double xdiff = x - (int) x;
            double ydiff = y - (int) y;

            int xf = (int) Math.floor(x);
            int xc = (int) Math.ceil(x);
            int yf = (int) Math.floor(y);
            int yc = (int) Math.ceil(y);

            double out = inputArray[xf][yf] * (1 - xdiff) * (1 - ydiff)
                    + inputArray[xc][yf] * xdiff * (1 - ydiff)
                    + inputArray[xf][yc] * (1 - xdiff) * ydiff
                    + inputArray[xc][yc] * xdiff * ydiff;

            outputArray[i][j] = (int) out;
        }
    }

    long elapsed = System.currentTimeMillis() - start;
    System.out.println("Time used: " + elapsed);

在得出该代码之后，我想知道不为下限值和上限值创建 4 个临时变量而是直接在数组索引中计算它们是否会更快。所以我这样修改:

    long start = System.currentTimeMillis();

    for (int i = 0; i < newWidth; i++) {
        for (int j = 0; j < newHeight; j++) {

            double x = i * scaleX;
            double y = j * scaleY;

            double xdiff = x - (int) x;
            double ydiff = y - (int) y;

            double out = inputArray[(int) Math.floor(x)][(int) Math.floor(y)] * (1 - xdiff) * (1 - ydiff)
                    + inputArray[(int) Math.ceil(x)][(int) Math.floor(y)] * xdiff * (1 - ydiff)
                    + inputArray[(int) Math.floor(x)][(int) Math.ceil(y)] * (1 - xdiff) * ydiff
                    + inputArray[(int) Math.ceil(x)][(int) Math.ceil(y)] * xdiff * ydiff;

            outputArray[i][j] = (int) out;
        }
    }

    long elapsed = System.currentTimeMillis() - start;
    System.out.println("Time used: " + elapsed);

我原以为后者会更快(因为您不必写入一个临时变量然后访问它)但事实证明后者至少比前者慢 2.5 倍。使用的测试用例是 1024x768 img 的 3 倍缩放。

原代码:使用时间:812后来的代码:使用时间:2140

那么造成时差的原因是什么？

Answer 1

您在后面的代码中有 8 个 Math.floor()/Math.ceil() 计算而不是 4 个。这就是问题所在。

计算比为局部变量分配空间慢得多。