C语言实现多线程的归并排序

问题

利用多线程实现归并排序。归并排序是分治算法的代表，适合改写为多线程。

构造数据

数据规模为 2 ∗ 1 0 6 2*10^6 2∗106,整数。 a r r a y _ l e n g t h = 2 ∗ 1 0 6 array\_length = 2*10^6 array_length=2∗106

    srand((int)time(NULL));for (int i = 0; i < array_length; ++i) {a[i] = rand();}

计时

利用 < s y s / t i m e . h > <sys/time.h> <sys/time.h>内部提供的 g e t n u m o f d a y ( ) getnumofday() getnumofday()函数，可以精确到微妙级别，但我最后输出是毫秒级别，所以需要转换单位。代码如下：

struct timeval tbegin, tend;gettimeofday(&tbegin, NULL);int arg[2];arg[0] = 0;arg[1] = array_length;pthread_t  tid;numofThread = 1;pthread_create(&tid, NULL, merge_sort, arg);pthread_join(tid, NULL);gettimeofday(&tend, NULL);if (flag == 1) {printf("The number of thread that I use : %d\n", maxThreadNumber);     }

归并排序

我需要控制好线程的使用个数，因此要统计当前正在运行的线程个数。为了避免潜在的bug，我并没有用到多少指针的东西，基本都是全局数组全局变量。归并排序其实有两个函数， m e r g e _ s o r t merge\_sort merge_sort函数是递归实现分治的基础。我在该函数的单线程版本上进行了些许改动。

代码如下：

void merge_sort(void* arg){int *argu = (int*)arg;int left = argu[0];int right = argu[1];int mid = (left + right) >> 1;int arg1[2];int arg2[2];arg1[0] = left;arg1[1] = mid;arg2[0] = mid + 1;arg2[1] = right;if (left >= right) {return;}pthread_t t2;pthread_t t1;if (numofThread == maxThreadNumber) {flag = 1;}if (numofThread < maxThreadNumber) {numofThread += 1; pthread_create(&t1, NULL, merge_sort, arg1);pthread_join(t1, NULL); pthread_exit(NULL);numofThread -= 1;}else {merge_sort(arg1);}if (numofThread < maxThreadNumber) {numofThread += 1;pthread_create(&t2, NULL, merge_sort, arg2); pthread_join(t2, NULL);pthread_exit(NULL);numofThread -= 1;}else {merge_sort(arg2);}merge(left, right);}

完整代码

#include <stdio.h>
#include <pthread.h>
#include <stdlib.h>
#include <string.h>
#include <sys/time.h>
#include <time.h>
#define DEBUG 0
#define array_length 2000000int a[array_length+5];
int numofThread = 0;
int maxThreadNumber = 21;
int flag = 0;void merge(int left, int right){int mid = (left + right) >> 1;int size1 = mid - left + 1;int size2 = right - mid;int t1[size1];int t2[size2];memcpy(t1, a+left, sizeof(int) * (mid-left+1));memcpy(t2, a+mid+1, sizeof(int) * (right-mid));int i = 0, j = 0;int k = left;while (i < size1 && j < size2) {if (t1[i] <= t2[j]) {a[k] = t1[i];i++;}else {a[k] = t2[j];j++;}k++;}while (i < size1) {a[k] = t1[i];k++;i++;}while (j < size2) {a[k] = t2[j];j++;k++;}
}void merge_sort(void* arg){int *argu = (int*)arg;int left = argu[0];int right = argu[1];int mid = (left + right) >> 1;int arg1[2];int arg2[2];arg1[0] = left;arg1[1] = mid;arg2[0] = mid + 1;arg2[1] = right;if (left >= right) {return;}pthread_t t2;pthread_t t1;if (numofThread == maxThreadNumber) {flag = 1;}if (numofThread < maxThreadNumber) {numofThread += 1; pthread_create(&t1, NULL, merge_sort, arg1);pthread_join(t1, NULL); pthread_exit(NULL);numofThread -= 1;}else {merge_sort(arg1);}if (numofThread < maxThreadNumber) {numofThread += 1;pthread_create(&t2, NULL, merge_sort, arg2); pthread_join(t2, NULL);pthread_exit(NULL);numofThread -= 1;}else {merge_sort(arg2);}merge(left, right);}void createData(){srand((int)time(NULL));for (int i = 0; i < array_length; ++i) {a[i] = rand();}
}int main(){createData();struct timeval tbegin, tend;gettimeofday(&tbegin, NULL);int arg[2];arg[0] = 0;arg[1] = array_length;pthread_t  tid;numofThread = 1;pthread_create(&tid, NULL, merge_sort, arg);pthread_join(tid, NULL);gettimeofday(&tend, NULL);if (flag == 1) {printf("The number of thread that I use : %d\n", maxThreadNumber);     }printf("The running time is %d millisecond\n",(tend.tv_usec - tbegin.tv_usec)/1000);#if DEBUG == 1for (int i = 0; i < array_length; ++i) {printf("%d\n",a[i]);}
#endifreturn 0;
}

性能分析

下面比较各种线程数量，我的程序的运行效果。

从这4个不同的线程数量可以对比看出，线程数越多，程序运行时间是越短的。单线程运行效率最低，多线程每次线程数翻倍，运行时间大大缩短。而且并不是只缩短一半，缩短幅度大大提高。

总结

归并排序并不难，改写为多线程版本花费了不少时间在调试代码上。说明我对多线程的掌握还不算到位。但是通过这次练习，我有了更深刻的了解。