递归数组方法在Java中返回0

Question

I have taken the tip I got from my last question and went deep into debugging my code. I'm still trying to make a method that takes two multidimensional arrays n*n and multiplies them, as if they were matrices. But I have encountered an odd obstacle. This is my code:

public class subMatrix {

    public int[][] divcon(int[][] a, int[][] b, int sub) {

        int[][] matrixC = new int[a.length][a.length];

        if (sub == 1) {
            matrixC[0][0] = a[0][0] * b[0][0];

        } else {
            sub = sub / 2;

            int[][] smalla11 = new int[sub][sub];
            int[][] smalla12 = new int[sub][sub];
            int[][] smalla21 = new int[sub][sub];
            int[][] smalla22 = new int[sub][sub];
            int[][] smallb11 = new int[sub][sub];
            int[][] smallb12 = new int[sub][sub];
            int[][] smallb21 = new int[sub][sub];
            int[][] smallb22 = new int[sub][sub];
            int[][] smallc11 = new int[sub][sub];
            int[][] smallc12 = new int[sub][sub];
            int[][] smallc21 = new int[sub][sub];
            int[][] smallc22 = new int[sub][sub];

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

                    smalla11[i][j] = a[i][j];
                    smalla12[i][j] = a[sub + i][j];
                    smalla21[i][j] = a[i][sub + j];
                    smalla22[i][j] = a[sub + i][sub + j];

                    smallb11[i][j] = b[i][j];
                    smallb12[i][j] = b[sub + i][j];
                    smallb21[i][j] = b[i][sub + j];
                    smallb22[i][j] = b[sub + i][sub + j];

                }
            }
            smallc11 = addMatrix(divcon(smalla11, smallb11, sub), divcon(smalla12, smallb21, sub), sub);
            smallc12 = addMatrix(divcon(smalla11, smallb12, sub), divcon(smalla12, smallb12, sub), sub);
            smallc21 = addMatrix(divcon(smalla21, smallb11, sub), divcon(smalla22, smallb21, sub), sub);
            smallc22 = addMatrix(divcon(smalla21, smallb12, sub), divcon(smalla22, smallb22, sub), sub);
        }
        return matrixC;
    }

    public int[][] addMatrix(int[][] aadd, int[][] badd, int size) {
        int[][] c = new int[size][size];

        for (int d = 0; d < size; d++) {
            for (int e = 0; e < size; e++) {

                c[d][e] = aadd[d][e] + badd[d][e];
            }
        }
        return c;
    }
}

The matrices a and b are turned into quarters. I have debugged my code with two matrice inputs 4x4 with the value 2 in each field. Until a certain point, things are going the right way. But if I add a System.out.println() to see the return values from smallc11 , 12, 21 and 22, it sometimes shows as 0. How come? I'm never inputting any 0, and the arrays should not go lower than 1 in length which is 2.

Answer 1

You are returning matrixC but you are not altering it anywhere, so it will be full of 0 .

You can copy the results from all your sub-matrixes, but the simpler thing to do is not create them in the first place. Just add all the elements of the two original matrixes together will be simpler and faster.

Something like

public double[][] multiply(double[][] a, double[][] b) {
     assert a[0].length == b.length;
     double[][]c = new double[a.length][b[0].length];
     for(int i = 0, len1 = a.length; i < len1; a++) {
         for(int j = 0, len2 = b[0].length; j < len2; j++) {
              double sum = 0.0;
              for(int k = 0, len3 = b.length; k < len3; k++)
                  sum += a[i][k] * b[k][j];
              c[i][j] = sum;
         }
     }
}

The reason this is much faster is the cost of creating new objects and copying values exceeds the cost of the * and + .

You can optimise this further by transposing b first to improve cache friendliness.