## Reproduce
Run the reproducer with:
  -1) java TestSuperWordOverunrolling
  -2) java -XX:-SuperWordLoopUnrollAnalysis TestSuperWordOverunrolling
---------------------------------
public class TestSuperWordOverunrolling {

    public static void main(String[] args) {
        double sum = 0.0;
        long start = System.currentTimeMillis();
        for (int i = 0; i < 50000; i++) {
            sum += execute(256);
        }
        long end = System.currentTimeMillis();
        System.out.println("sum = " + sum + "; time = " + (end - start) + "ms");
    }

    public static double execute(int num_iterations) {
        int M = 63;
        byte[][] G = new byte[M][M];

        int Mm1 = M-1;
        for (int p = 0; p < num_iterations; p++) {
            for (int i = 1; i < Mm1; i++) {
                for (int j = 1; j < Mm1; j++)
                    G[i][j] = G[i-1][j];
            }
        }

        return G[3][2];
    }
}
---------------------------------

## Symptom
1) java TestSuperWordOverunrolling
---------------------------------
sum = 0.0; time = 9360ms
sum = 0.0; time = 9345ms
sum = 0.0; time = 9376ms
sum = 0.0; time = 9389ms
---------------------------------

2) java -XX:-SuperWordLoopUnrollAnalysis TestSuperWordOverunrolling
---------------------------------
sum = 0.0; time = 5564ms
sum = 0.0; time = 5575ms
sum = 0.0; time = 5520ms
sum = 0.0; time = 5552ms
---------------------------------

## Analysis
The performance drop was caused by over loop unrolling with SuperWordLoopUnrollAnalysis.
For this reproducer, the loop was unrolled by 16, which was bad for the performance.