Mystery of the JVM

O4Matrix.java

//        Round 1: unoptimized version
//        why is here first cycle is too short and rest are too long?
//        1.639
//        3.333
//        3.298
//        3.289
//
//        Round 2: unoptimized version (again!)
//        why each cycle takes longer than first cycle of first round?
//        and less than others?
//        1.975
//        1.977
//        1.977
//        1.983
//        1.976
//        1.975
//        1.974
//        1.974
//        1.983
//        1.982
//
//        Round 3: optimized version
//        1.523
//        1.522
//        1.531
//        1.526
//        1.538
//        1.532
//        1.522
//        1.52
//        1.52
//        1.52
public class O4Matrix {

    public static void main(String[] args) {
        float[] array1 = new float[9];
        float[] array2 = new float[9];
        float[] array3 = new float[9];

        System.out.println("Round 1: unoptimized version");
        System.out.println("  why is here first cycle is too short and rest are too long?");
        for (int k = 0; k < 4; k++) {
            long start = System.currentTimeMillis();
            for (int i = 0; i < 100000000; i++) O4Matrix.multiply33(array1, array2, array3);
            System.out.println((System.currentTimeMillis() - start) / 1e3f);
        }

        System.out.println();
        System.out.println("Round 2: unoptimized version (again!)");
        System.out.println("  why each cycle takes longer than first cycle of first round?");
        System.out.println("  and less than others?");
        for (int k = 0; k < 10; k++) {
            long start = System.currentTimeMillis();
            for (int i = 0; i < 100000000; i++) O4Matrix.multiply33(array1, array2, array3);
            System.out.println((System.currentTimeMillis() - start) / 1e3f);
        }

        System.out.println();
        System.out.println("Round 3: optimized version");
        for (int k = 0; k < 10; k++) {
            long start = System.currentTimeMillis();
            for (int i = 0; i < 100000000; i++) O4Matrix.multiply33b(array1, array2, array3);
            System.out.println((System.currentTimeMillis() - start) / 1e3f);
        }
    }

    public O4Matrix() {
    }

    public static void set(int ww, float[] data, int row, int column, float d) {
        data[(row - 1) * ww + column - 1] = d;
    }

    public static float get(int ww, float[] data, int row, int column) {
        return data[(row - 1) * ww + column - 1];
    }

    public static void multiply(int w, int h, float[] ab, float[] a, float[] b) {
        for (int i = 1; i <= w; i++) {
            for (int j = 1; j <= h; j++) {
                float r = 0;
                for (int k = 1; k <= h; k++) r += get(w, a, i, k) * get(w, b, k, j);
                set(w, ab, i, j, r);
            }
        }
    }

    public static void multiply33(float[] ab, float[] a, float[] b) {
        multiply(3, 3, ab, a, b);
    }

    public static void multiply33b(float[] ab, float[] a, float[] b) {
        ab[0] = a[0] * b[0] + a[1] * b[3] + a[2] * b[6];
        ab[1] = a[0] * b[1] + a[1] * b[4] + a[2] * b[7];
        ab[2] = a[0] * b[2] + a[1] * b[5] + a[2] * b[8];
        ab[3] = a[3] * b[0] + a[4] * b[3] + a[5] * b[6];
        ab[4] = a[3] * b[1] + a[4] * b[4] + a[5] * b[7];
        ab[5] = a[3] * b[2] + a[4] * b[5] + a[5] * b[8];
        ab[6] = a[6] * b[0] + a[7] * b[3] + a[8] * b[6];
        ab[7] = a[6] * b[1] + a[7] * b[4] + a[8] * b[7];
        ab[8] = a[6] * b[2] + a[7] * b[5] + a[8] * b[8];
    }

}