OpenMP程序 for 循环并行的效率

#pragma omp parallel for 这条语句是用来指定后面的for循环语句变成并行执行的,将for循环里的语句变成并行执行后效率会不会提高呢?还是测试一 下吧，测试的时候，循环最好大一些，增加计算量，不然程序耗时太短，很难区分。

#include <omp.h>#include <stdio.h>#include <sys/time.h>void test() {int a = 0;for(int i = 0; i < 10000000; i++) {a = i + 1;}}int main() {int nthrds = 4;omp_set_num_threads(nthrds);timeval tStart,cTime;long long tPassed = 0;gettimeofday(&tStart, 0);for (int i = 0; i < 100; i++) {test();}gettimeofday(&cTime, 0);cTime.tv_sec -= tStart.tv_sec; cTime.tv_usec -= tStart.tv_usec;tPassed = 1000000LL * cTime.tv_sec + cTime.tv_usec; tPassed /= 1000;printf("Time = %lld/n", tPassed);gettimeofday(&tStart, 0);#pragma omp parallel forfor (int i = 0; i < 100; i++) {test();}gettimeofday(&cTime, 0);cTime.tv_sec -= tStart.tv_sec;cTime.tv_usec -= tStart.tv_usec;tPassed = 1000000LL * cTime.tv_sec + cTime.tv_usec; tPassed /= 1000;printf("Time = %lld/n", tPassed);return 1;}

 result:

[root@c0108 zlt]# g++ omp.c -o omp -fopenmp[root@c0108 zlt]# ./ompTime = 2859Time = 718[root@c0108 zlt]#  

在四核的机器上，开了四个线程，加速比基本上是4，果然不错。

 

 

此外，BS一下clock()函数，我向来觉得它一无是处。。。

#include <omp.h>#include <stdio.h>#include <time.h>void test() { int a = 0; for(int i = 0; i < 10000000; i++) { a = i + 1; }}int main() { int nthrds = 4; omp_set_num_threads(nthrds); long start = clock(); for (int i = 0; i < 100; i++) { test(); } long end = clock(); printf("Time = %ld/n", end - start); start = clock(); #pragma omp parallel for for (int i = 0; i < 100; i++) { test(); } end = clock(); printf("Time = %ld/n", end - start); return 1;} 

 

结果：

[root@c0108 zlt]# g++ omp.c -o omp -fopenmp[root@c0108 zlt]# ./ompTime = 2860000Time = 2860000[root@c0108 zlt]#  

clock有三个问题:
1)如果超过一个小时,将要导致溢出.
2)函数clock没有考虑CPU被子进程使用的情况.
3)也不能区分用户空间和内核空间.