第九章 通用核心回顾

• 创建线程组:
  • shared(list)
  • private(list)
  • firstprivate(list)
  • default(none)
  • reduction(operator:list)

#pragma omp parallel [clause[,clause]...]
{
    //body of parallel
}

• 共享工作循环构造:
  • private(list)
  • firstprivate(list)
  • nowait
  • reduction(operator:list)
  • schedule(kind[,chunk_size])

#pragma omp for [clause[,clause]...]
    //for-loop

• single构造:
  • private(list)
  • firstprivate(list)
  • nowait

#pragma omp single [clause[,clause]...]
{
    //body of single
}

• 组合式并行共享工作循环构造:
  • shared(list)
  • private(list)
  • firstprivate(list)
  • reduction(operator:list)
  • schedule(kind[,chunk_size])

#pragma omp parallel for [clause[,clause]...]
    //for-loop

• 创建显式任务:
  • shared(list)
  • private(list)
  • firstprivate(list)
  • default(none)

#pragma omp task [clause[,clause]...]
{
    //body of task
}

• 任务等待指令:

#pragma omp taskwait

• 栅栏指令:

#pragma omp barrier

• 临界区构造:

#pragma omp critical
{
    //body of critical
}

• 常用的数据环境子句:

  • shared(list)
  • private(list)
  • firstprivate(list)

• 归约子句:

  • reduction(operator:list)

• 设置默认并行区域请求线程数的环境变量:

OMP_NUM_THREADS list

• 设置后继并行区域的默认请求线程数:

void omp_set_num_threads(int num_threads);

• 返回最内围并行区域当前组的线程数:

int omp_get_num_threads(void);

• 返回最内围并行区域当前组内调用线程的线程号:

int omp_get_thread_num(void);

• 返回经过的墙钟时间:

double omp_get_wtime(void);