OpenMP有三种常见的加锁操作：

critical是OpenMP的指令，它规定其后的代码为临界块，任何时候只允许一个线程访问；

omp_set_lock是OpenMP的库函数，要跟omp_unset_lock一起使用；

atomic也是指令，其后的内存位置将会原子更新

例子：

int main()
{
    int i, nVar = 0, n = 5000000;
    omp_set_num_threads(4);
    clock_t t1, t2;

    //critical
    t1 = clock();
#pragma omp parallel for shared(nVar)
    for (i = 0; i < n; i++)
    {
#pragma omp critical
        {
            nVar += 1;
        }
    }

    t2 = clock();

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

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

    t1 = clock();
    omp_lock_t mylock;
    omp_init_lock(&mylock);
#pragma omp  parallel for shared(nVar)
    for (i = 0; i < n; i++)
        {
        omp_set_lock(&mylock);
        nVar += 1;
        omp_unset_lock(&mylock);
    }
    omp_destroy_lock(&mylock);
    t2 = clock();

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

    //atomic
    nVar = 0;
    t1 = clock();

#pragma omp parallel for shared(nVar)
    for (i = 0; i < n; i++)
    {
#pragma omp atomic
        nVar += 1;
    }
    t2 = clock();
    printf("atomic nVar:%d\n", nVar);
    printf("atomic time: %ld ms\n", t2 - t1);
    printf("---------------\n");

    //normal
    nVar = 0;
    t1 = clock();
    for (i = 0; i < n; i++)
    {
        nVar += 1;
    }
    t2 = clock();

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

    system("pause");
    return 0;
}

在这个任务里是不并行比并行性能好啊。如果要并行的话，加锁的时候还是atomic的性能好一些。