利用 openmp 实现在intel多核系统上 基本排序算法性能比较

Note：如果您觉得这篇文章对您有用或者对您有一点点帮助，麻烦您为她投票，谢谢.

本文章因为参加投票，请大家转载时说明，谢谢。

投票地址：http://intel.csdn.net/multicoreblog/show.aspx?page=0 （请投对文章：），谢谢）

本篇文章通过使用 openMp 实现对基本排序算法的性能比较：

文章很简单，但是很实用，分别演示了如何使用 openMP 进行多核多线程程序开发，可以作为入门级朋友的

学习教材。如果想正确运行程序，您需要创建每个文件，同时命名一定要合适，否则您需要自己去重写

后面的 makefile 文件。

 

sorh.h 头文件

//sort.h head file#include <stdio.h>#include <time.h>#include <stdlib.h>#include <omp.h>#define NUM_THREADS 12typedef float KeyType;typedef int KeyInt;typedef struct LNode{KeyType key;struct LNode* next;}LNode;

 

sort.c 文件 包含main 函数，相当于一个 client 实现对各个排序算法的并行调用。

这里采用了 openMp 指导语句 #pragma omp sections 实现结构块化，即把各个结构块划分到各自的线程上。实际上是一个 任务分配区（work-sharing sections）的概念。通过 #pragma omp parallel

#pragma omp sections 指导OpenMp编译器和运行时库将应用程序中标示出的结构化块分配到用于执行并行区域的一

组线程上。这段例子还是非常有用的。

# sort.c#include "sort.h"/**** The program need assitant Functions ****/KeyType* randomCreat(int N);void DisPlay(int N, KeyType *p);int GetExponent(int N);KeyInt* Dlta(int exponent); //build Dlta array for ShellSortvoid FindPrimes(int start, int end); //for Dlta Maybe this is a badlly methodint TestForPrime(int val); //for Dlta void myKeyboardFunc( unsigned char key);/**** Bubble Sort Functions in Bubble.c**/KeyType* BubbleArgorithm(int N, KeyType *p);/****Merge Sort Functions in Merge.c**/KeyType* MMSort(int N, KeyType* p);void MSort(KeyType *r1, KeyType *r2, int left, int m, int r);void MergeSort(KeyType *r1, KeyType *r2, int left, int right);/****InsertSort Functions in InsertSort.c**/KeyType* InsertSort(int N, KeyType* p);/****ShellSort Functionsin ShellSort.c**/KeyType* ShellSort(KeyType* p, KeyInt* dlta, int t, int N);void ShellInsert(KeyType* p, int dk, int N);/****QuickSort Functionsin QuickSort.c**/KeyType* QuickSort(int N, KeyType* p);KeyType Partition(KeyType* p, int low, int high);void Qsort(KeyType* p, int low, int high);/****BucketSort Functionsin BucketSort.c**/LNode** Bucket(int N); //create Bucket according to NKeyType* BucketSort(LNode** bucket, KeyType* p, int N);int main(int argc, char *argv[]){ clock_t start, stop; char any = 'c';KeyType *p0, *p1, *p2, *p3, *p4, *p5;LNode** bucket; //for BucketSort B【】int N, i; //The size is sorted. if(argc < 2){printf("Enter random size N=");scanf_s("%d",&N);}else{N = atoi(argv[1]);}/**** create sorted random copy**/p0 = randomCreat(N); //p0 for Merge-sortp1 = malloc(N*sizeof(KeyType));//p1 for Bubble-sortp2 = malloc(N*sizeof(KeyType));//p2 for Insert-sortp3 = malloc(N*sizeof(KeyType));//p3 for Shell-sortp4 = malloc(N*sizeof(KeyType));//p4 for Quick-sortp5 = malloc(N*sizeof(KeyType));//p5 for Bucket-sortfor( i=0; i<N; i++){p1[i] = p2[i] = p3[i] = p4[i] = p5[i] = p0[i];}//DisPlay(N, p0);/**** Start testing method of sort.** Use OpenMP create threads deal with methods ****/#pragma omp parallel{#pragma omp sections nowait{/**** Merge sort ** **/#pragma omp section{start = clock();p0 = MMSort(N, p0);stop = clock(); printf("/b/b/b/b/bThe MergeSort elapsed time = %g seconds, The threads num is %d/n ", (double)(stop - start) / CLOCKS_PER_SEC,(int)omp_get_thread_num());//DisPlay(N, p0);free(p0);}/**** Bubble sort**/#pragma omp section{start = clock();p1 = BubbleArgorithm(N, p1);stop = clock(); printf("/b/b/b/bThe BubbleSort elapsed time = %g seconds, The threads num is %d/n ", (double)(stop - start) / CLOCKS_PER_SEC,(int)omp_get_thread_num());//DisPlay(N, p1);free(p1);}/**** Insert-sort **/#pragma omp section{ start = clock(); p2 = InsertSort(N, p2); stop = clock(); printf("/b/b/b/bThe InsertSort elapsed time = %g seconds, The threads num is %d/n ", (double)(stop - start) / CLOCKS_PER_SEC,(int)omp_get_thread_num());// DisPlay(N, p2); free(p2);}/**** Shell-sort **/#pragma omp section{ start = clock(); //According to N made dlta size and t size p3 = ShellSort(p3, Dlta(GetExponent(N)), GetExponent(N), N); stop = clock(); printf("/b/b/b/bThe Shellsort elapsed time = %g seconds, The threads num is %d/n ", (double)(stop - start) / CLOCKS_PER_SEC,(int)omp_get_thread_num()); //DisPlay(N, p3); free(p3);}/**** Quick-sort **/#pragma omp section{start = clock();p4 = QuickSort(N, p4);stop = clock(); printf("/b/b/b/bThe QuickSort elapsed time = %g seconds, The threads num is %d/n ", (double)(stop - start) / CLOCKS_PER_SEC,(int)omp_get_thread_num());//DisPlay(N, p4);free(p4);}/**** BucketSort **/#pragma omp section{start = clock();bucket = Bucket(N);p5 = (KeyType*)FindBucket(N, p5, bucket);stop = clock(); printf("/b/b/b/bThe BucketSort elapsed time = %g seconds, The threads num is %d/n ", (double)(stop - start) / CLOCKS_PER_SEC,(int)omp_get_thread_num());//DisPlay(N, p5);free(p5);}}}printf("/b/b/b/b/b/bPlease press /"q/" to quit program ..................!!!/n"); // any = getchar(); while(1){ if(getchar()=='q') return 0;}}

// bubble.c 实现冒泡排序#include "sort.h"/*** bubbel argorithm * 第一次循环将最大的数放到最后，后面循环依次类推。*/KeyType* BubbleArgorithm(int N, KeyType *p){ int i, j; KeyType temp; for(i=0; i<N; i++) for(j=N-1; j>i; j--) if(p[j]<p[j-1]) { temp = p[j-1]; p[j-1] = p[j]; p[j] = temp; } return (p);}

 

buctet.c 文件实现 桶排序

#include "sort.h"/****According to N create B[N] for bucket**/LNode** Bucket(int N){int i=0;LNode *q, **p;p = malloc(N*sizeof(LNode*));//initialization p using OpenMP#pragma omp parallel for private(i) for(i=0; i<N; i++){q = malloc(sizeof(LNode*));q -> key = 0.0;q -> next = NULL;p[i] = q;}return (p);}//Function is insert a Node to a order LinkListKeyType* FindBucket(int N, KeyType* p, LNode** q){int i, j=0;long tmp;LNode *ele;KeyType* finall;for(i=0; i<N; i++){tmp = (long)(p[i] * N);ele = malloc(sizeof(LNode*));ele->key = p[i];ele->next = NULL;q[tmp]->next = (LNode*)LinkInsert(q[tmp]->next, ele);}finall = malloc(N*sizeof(KeyType)); for(i=0; i<N; i++){while(q[i]->next){finall[j]= q[i] -> next -> key;q[i] ->next = q[i] -> next->next;j++;}} return (finall);}//Function is insert a Node to a order LinkListLNode* LinkInsert(LNode *LinkList, LNode *p){LNode *current, *head;current = LinkList;head = NULL; while((current != NULL) && (current->key < p->key)){head = current;current = current->next;}p->next = current;if( head == NULL ){current = p;return current;}else{head->next = p;return head;}}

 

InsertSort.c

#include "sort.h"/**** @function: Insert sort** @**/KeyType* InsertSort(int N, KeyType* p){ int i, j; KeyType tmp; for(i=0; i<N-1; ++i){ if(p[i+1] < p[i]){ tmp = p[i+1]; p[i+1] = p[i]; for(j = i; tmp < p[j]; --j) p[j+1] = p[j]; p[j+1] = tmp;}} return (p);}

 

Merge.c

#include "sort.h"/****** mSort ** ** i is left j is middle k is left***/void MSort(KeyType *r1, KeyType *r2, int left, int m, int r){int i = left, j = m + 1, k = left;while((i <= m)&&(j <= r)){if(r1[i] <= r1[j]){r2[k++] = r1[i++];}else{r2[k++] = r1[j++];}}//end while //将剩余的人r1 内容复制到 r2 中while(i <= m){r2[k++] = r1[i++];} //将剩余的人r1 内容复制到 r2 中 while(j <= r){r2[k++] = r1[j++];}//修改r1内容 for(k=1; k <= r; k++){r1[k] = r2[k];}}/** * MergeSort*/void MergeSort(KeyType *r1, KeyType *r2, int left, int right){int middle;if(left < right){middle = ((left + right) >> 1);#pragma omp parallel{#pragma omp sections nowait{#pragma omp sectionMergeSort(r1, r2, left, middle);#pragma omp sectionMergeSort(r1, r2, middle + 1, right );}}MSort(r1, r2, left, middle, right);}}KeyType* MMSort(int N, KeyType *r){KeyType *tmp;if((tmp = malloc(N * sizeof(KeyType)))){MergeSort(r, tmp, 0, N-1);free(tmp);return (r);}else{printf("malloc is failed !!!");return (r);}}//因为参与投票，文章版权所有请不要转发。 唐炜

 

QuickSort.c  快速排序可以实用两个线程分别对各个待排序队列实现递归排序

#include "sort.h"/****@function: QuickSort**@author: Tang Wei****/KeyType Partition(KeyType* p, int low, int high){KeyType tmp = p[low]; //pivotkeywhile(low < high){while((low < high) && (p[high] >= tmp)) --high;p[low] = p[high];while((low < high) && (p[low] <= tmp))++low;p[high] = p[low];}p[low] = tmp;return low;}void Qsort(KeyType* p, int low, int high){int pivotLocation;if(low < high){pivotLocation = Partition(p, low, high);#pragma omp parallel {#pragma omp sections nowait{#pragma omp sectionQsort(p, low, pivotLocation-1);#pragma omp sectionQsort(p, pivotLocation+1, high);}}}}KeyType* QuickSort(int N, KeyType* p){Qsort(p, 0, N-1);return(p);}

 

rand.c

 

#include "sort.h"/*** 生成随即序列**/KeyType* randomCreat(int N){int i=0;KeyType *p;p = malloc(N * sizeof(KeyType));for(i=0; i<N; i++)p[i] = rand()/(RAND_MAX+1.0); return (p);}void DisPlay(int N, KeyType *p){while(N>0)printf("/b/b/b/b/b/b/bp[%d]:=%f/n",N, p[--N]);}void myKeyboardFunc( unsigned char key){switch ( key ) { case 'q': //****bobc // exit program exit(0); //****bobc}}

 

ShellSort.c

 

#include "sort.h"static int gProgress = 0, gPrimesFound = 0;long globalPrimes[20];/****@function: Shell Sort**/void ShellInsert(KeyType* p, int dk, int N){int i, j;KeyType tmp;for(i = dk + 1; i < N; ++i){if(p[i] < p[i - dk]){tmp = p[i];for(j = i - dk; j>0 && (tmp < p[j]); j-=dk)p[j + dk] = p[j];p[j + dk] = tmp;}}}KeyType* ShellSort(KeyType* p, KeyInt* dlta, int t, int N){int k;for(k = 0; k < t; ++k)ShellInsert(p, dlta[k], N);free(dlta);return (p);}/****** First, calculate t for dlat[] ** Second, insert dlta[] with primes numbers** Third, return dlta*/int GetExponent(int N){int exponent = (int)(log10(N)/log10(2));return exponent;}int* Dlta(int exponent){//int exponent = (int)(log10(N)/log10(2));KeyInt* p;int i;p = malloc(exponent*sizeof(KeyInt)); //assign space for dlta;FindPrimes(1, 100);for(i=0; i<exponent-1; i++){//printf("i: =%d, exponent-2-i=%d, globalPrimes[exponent-1-i]:=%d/n", i,exponent-1-i, globalPrimes[exponent-i]);p[i] = globalPrimes[exponent-i];}p[exponent-1] = 1;return (p);}/** 这是采用分别取线程号的方法实现并行*/void FindPrimes(int start, int end){int i, k;omp_set_num_threads(NUM_THREADS);if (start <= 2) globalPrimes[gPrimesFound++] = 2;#pragma omp parallel for private(i)for(i = start; i <= end; i += 2 ){if( TestForPrime(i) ){//这里需要设置临界区否则 globalPrimes[] 里面的内容不合适#pragma omp criticalglobalPrimes[gPrimesFound++] = i;//printf("prime : %d/n", globalPrimes[gPrimesFound]);}}}int TestForPrime(int val){ int limit, factor = 3; limit = (long)(sqrtf((float)val)+0.5f); while( (factor <= limit) && (val % factor)) factor ++; return (factor > limit);}//因为参与投票，文章版权所有请不要转发。 唐炜

 

MAKEFILE.mak  我这里建立了一个 program 包，将所有的文件放到这个目录里面，如果您不一样的话，请修改

makefile文件  CPP_PROJ = $(CFLAGS) /Qopenmp -c  ../program 内容 /Qopenmp 为打开Qopenmp选项，

编译器支持openMP，  ../program 为目录

 

############################################################################################# You must use Intel C++ Compiler ###################### C++ Build Environment for applications running on IA-32 ###################### command line> nmake /f MAKEFILE.mak ###################### #################################################################################################### Compiler, tools and optionsCPP=icl.exe!IF "$(CPP)" == "icl.exe"LINK=xilink.exe!ELSELINK=link.exe!ENDIFCF = -O3LF = CFLAGS = $(CF) LFLAGS = $(LF)CPP_PROJ = $(CFLAGS) /Qopenmp -c ../programSRCS = *.cppOBJS = *.objTARGET = sort.exe####### FilesSOURCES = sort.c rand.c Merge.c Bubble.c InsertSort.c Shellsort.c QuickSort.c BucketSort.cOBJECTS = sort.obj rand.obj Merge.obj Bubble.obj InsertSort.obj ShellSort.obj QuickSort.obj BucketSort.obj####### Implicit rules.SUFFIXES: .cpp .cxx .cc .c.cpp.obj: $(CPP) -c $(CPP_PROJ) -Fo$@ $<.cxx.obj: $(CPP) -c $(CPP_PROJ) -Fo$@ $<.cc.obj: $(CPP) -c $(CPP_PROJ) -Fo$@ $<.c.obj: $(CPP) -c $(CPP_PROJ) -Fo$@ $<####### Build rulesall: $(TARGET)$(TARGET): $(OBJECTS) $(LIBS) $(LINK) $(LFLAGS) $(LDFLAGS) /OUT:$(TARGET) $(LIBS) $(OBJECTS) clean: -@erase *.obj -@erase *.bak -@erase *.exe####### Compile$(OBJECTS): $(SOURCES) $(CPP) -c $(CPP_PROJ) $(SOURCES)####### Libraries