OpenMP critical Lock() atomic 3种锁的比较

OpenMP critical Lock() atomic 3种锁的比较

//critical
//临界区在同一时间只能有一个线程执行它,其它线程要执行临界区则需要排队来执行它
//&//Lock() & Unlock()
//&atomic
//上面三者的耗时比比例大约系:7:3.5:1,原子操作最快,不过限制较前两者多…

int main()
{
         int i,nVar=0,n=1000000;
         omp_set_num_threads(4);
         clock_t t1,t2;

         //critical
         t1=clock();

#pragma omp parallel for shared(nVar) //schedule(static,1)
         for (i=0; i<n; i++)
         {
         #pragma omp critical //critical命令最终会被翻译成加锁和解锁的操作
             {
                nVar+=1;
                //printf("ThreadId: %d, i:%d, nVar:%d\n",omp_get_thread_num(),i,nVar);
             } //临界区在同一时间只能有一个线程执行它,其它线程要执行临界区则需要排队来执行它 ,这时等同于多个线程串行执行...
         }

         t2=clock();

         printf("nVar:%d\n",nVar);
         printf("time: %ld ms\n",t2-t1);

         printf("---------------\n");





         //Lock() & Unlock()
         nVar=0;

         t1=clock();

         omp_lock_t mylock;  // omp_lock_t mylock
         omp_init_lock(&mylock);//omp_init_lock(&mylock);
#pragmaomp parallel for shared(nVar) //schedule(static,2) //锁:主要用于对shared变量的操作,防止出现数据竞争
         for( i=0; i<n; i++)
         {
                   omp_set_lock(&mylock); //omp_set_lock(&mylock);
                   nVar+=1;
                   //printf("ThreadId: %d, i:%d, nVar:%d\n",omp_get_thread_num(),i,nVar);
                   omp_unset_lock(&mylock);//omp_unset_lock(&mylock);
         }
         omp_destroy_lock(&mylock);//omp_destroy_lock(&mylock);

         t2=clock();

         printf("nVar:%d\n",nVar);
         printf("time: %ld ms\n",t2-t1);

         printf("---------------\n");

         //atomic
         nVar=0;

         t1=clock();

#pragma omp parallel for shared(nVar)// schedule(static)
         for (i=0; i<n; i++)
         {
         #pragma omp atomic //atomic会被翻译成原子操作,比critical的锁操作要快,如果可以的话,尽量用原子操作代替锁操作
                   nVar+=1; //#pragma omp atomic
         //printf("ThreadId: %d, i:%d, nVar:%d\n",omp_get_thread_num(),i,nVar);
                    //临界区在同一时间只能有一个线程执行它,其它线程要执行临界区则需要排队来执行它 ,这时等同于多个线程串行执行...
         }

         t2=clock();

         printf("nVar:%d\n",nVar);
         printf("time: %ld ms\n",t2-t1);


         system("pause");
         return 0;
}

omp的Lock() & UnLock()的锁操作
//锁操作:
omp_lock_t mylock;

omp_init_lock(&mylock);
omp_destroy_lock(&mylock);

omp_set_lock(&mylock);
omp_unset_lock(&mylock);

int omp_test_lock(omp_lock_t * lock); //ture:成功上锁,false:解锁了

//嵌套锁操作:
omp_lock_t mylock;

omp_init_nest_lock(&mylock);
omp_destroy_nest_lock(&mylock);

omp_set_nest_lock(&mylock);
omp_unset_nest_lock(&mylock);

int omp_test_nest_lock(omp_lock_t * lock); //ture:成功上锁,false:解锁了