Question

我尝试使用fork join框架为填充浮点数的1D数组实现中值过滤器。我已设法按顺序执行此操作并且我的并行版本正在输出值，只是每次运行它输出不同的值时，例如当打印出已过滤数组的第一个元素时，我将得到-0.178966在再次运行之后，我将获得-0.168073，依此类推。我认为这可能是某种竞争条件，或者我在主要课堂上忽视某些事情或做错事。无论哪种方式，我希望你们中的一个人可以帮助我，我是并行编程的新手并且还没有完全理解它，所以请耐心等待。谢谢。（忽略主类中注释掉的代码）

我的主要课程和其他一些相关代码：

static final ForkJoinPool fjPool = new ForkJoinPool();

static void F(int fs,double[] nums, double[] ans)                              
{
   fjPool.invoke(new TFilter(fs,nums,0,nums.length,ans));
}

public static void main(String[] args)
{
    //note program requires user to enter file name, ask for filter size, and 
    //write to file with name being asked by program.
    //Scanner scn = new Scanner(System.in);
   // System.out.println("Enter name of file to be read from:");
    //String rfn = scn.nextLine();
    //System.out.println("Enter name of file to write to:");
    //String wfn = scn.nextLine();
    //System.out.println("Enter median filter size: ( must be odd and in range [3,21] )");
    //int fs = Integer.parseInt(scn.nextLine());

    double[] ss = read("sampleinputfile.txt"); //read fills array with input values 
    double[] a = new double[ss.length]; 
    F(3,ss,a);
    System.out.println(a[0]);

}

我的并行课程：

import java.util.*;
import java.util.concurrent.RecursiveAction;

public class TFilter extends RecursiveAction 
{
int lo;
int hi;
int fSize;
double [] arr;

double [] ans;

static final int SEQUENTIAL_CUTOFF = 500;

TFilter(int fs, double[] ar, int l, int h, double[] an)
{
    this.fSize = fs;
    this.arr = ar;
    this.lo = l;
    this.hi = h;
    this.ans = an;
}

public double median(double[] a)
{
    Arrays.sort(a);
    double m;
    if (a.length % 2 == 0)
    {
       m = ( a[a.length/2] + a[a.length/2 - 1])/2;
    }
    else
    {
       m = a[a.length/2];
    }

    return m;

}

public void MF()
{
    double[] Arr = new double[hi-lo];
    int cnt = 0;
    for (int i = lo; i < hi; i++)
    {
        Arr[cnt]  = arr[i];
        cnt++;
    }

    int m = (Arr.length-fSize)+1;
    List<Double> med = new ArrayList<>();
    double[] win = new double[fSize];

    for (int i = 0; i < m; i++)
    {
        int c = i;
        for (int j = 0; j < fSize; j++)
        {
            win[j] = Arr[c];
            c++;
        }
        med.add(median(win));

    }
    int x = (Arr.length-m)/2;

    for (int i = 0; i < x; i++) //first boundry points
    {
        ans[i] = Arr[i];
    }

    int c = 0;
    for (int i = x; i < Arr.length-x; i++)
    {
        ans[i] = med.get(c);
        c++;
    }
    for (int i = Arr.length-x; i < Arr.length; i++) //last boundry points
    {
        ans[i] = Arr[i];
    }

}


@Override
protected void compute() 
{                           
   if((hi-lo) < SEQUENTIAL_CUTOFF)
   {
       MF();  
   }
   else
   {
     TFilter left = new TFilter(fSize,arr,lo,(hi+lo)/2,ans);
     TFilter right = new TFilter(fSize,arr,(hi+lo)/2,hi,ans);

     left.fork();
     right.compute();
     left.join();

   }
}

}

Forkjoin框架每次运行时输出不同的值

0 个答案: