Question

我正在试图对用于字符串搜索的天真算法进行并行化。我创建了简单版本，然后尝试使用多个线程来加速它。

但是下面的代码会让它慢得多：

template<typename T> long unsigned int simple_paralel(const T * str1, unsigned long int str1_length, const T * str2, unsigned long int str2_length) {
  unsigned int long result = ~0;
  unsigned int long count = 0;

  unsigned long int in;
  unsigned int long top;
  #pragma omp parallel
  #pragma for ordered shared(result, count) private(in, top) firstprivate(str1, str2, str1_length, str2_length)
  for (top = 0; top < str1_length; top++) {
    in = 0;

    // & top + in < str1_length
    while (in < str2_length ) {
      if ( top + in >= str1_length)
        break;

      if ( str1[top+in] != str2[in] ) {
        break;
      }
      ++in;

      if( in == str2_length ) {
        // shared and we want to have the smallest index
        if( result >= top + 1 ) {
          result = top + 1;
        }
        count++;
      }
    }
  }
  return count;
}

我做错了什么？

Answer 1

所以我退后一步尝试了另一种方法：

template<typename T> long unsigned int simple_paralel(const T * str1, unsigned long int str1_length, const T * str2, unsigned long int str2_length, int threads) {
  count = 0;

  unsigned long int block = str1_length/threads;
  unsigned long int total = 0;
  unsigned long int result = 0;

  unsigned long int smallest = ~0;
  #pragma omp parallel for shared(count)
  for(int i=0; i<threads; i++)
  {
    unsigned long int res;
    unsigned long int start = i * block;
    if(i != threads -1 )
    {
      total += block;
      res = simple_p(str1 + start, block, str2, str2_length, false);
    } else {
      res = simple_p(str1 + start, str1_length - total, str2, str2_length);
    }

    if(res < result && res != 0 ){
      result = res;
    }
  }

  return count;
}

它按预期工作。但我仍然无法相信，我无法使用OpenMP 执行此操作。

字符串搜索 - 并行版本较慢

1 个答案: