C++ Array vs Vector性能测试说明

Question

为了量化C类数组和C++中的Vectors的性能差异,我编写了这个小程序.https://github.com/rajatkhanduja/Benchmarks/blob/master/C%2B%2B/vectorVsArray.cpp

为了比较它们的共同点,我决定对随机和顺序访问进行测试.我添加了迭代器,只是为了比较它们(但这不是问题所关注的).

对于具有7.7 GB RAM且阵列/矢量大小为100万的64位Linux计算机的结果如下: -

• 写入阵列所花费的时间.:12.0378毫秒
• 按顺序从数组中读取的时间.:2.48413 ms
• 从数组中随机读取的时间.:37.3931毫秒
• 写入动态数组所花费的时间.:11.7458毫秒
• 从动态数组顺序读取的时间.:2.85107毫秒
• 从动态数组中随机读取的时间.:36.0579毫秒
• 使用索引写入向量的时间.:11.3909毫秒
• 按顺序使用索引从矢量读取的时间.:4.09106毫秒
• 使用索引从矢量中读取的时间,随机.:39毫秒
• 使用迭代器写入向量的时间.:24.9949毫秒
• 使用迭代器从矢量读取的时间.:18.8049毫秒

向量的大小在初始化时设置而不是更改,因此不会调整向量的大小(程序中的断言有助于验证).时间不包括任何静态分配的数组,动态分配的数组或向量的初始化时间.

根据统计,写入Vector的时间小于数组的时间,但从向量读取的时间是数组的两倍.

差异很小,但有没有解释为什么会有性能差异？测试有问题吗？我希望两者都以相同的速度运行.该测试的重复显示了相同的趋势.

代码:

#include <vector>
#include <iostream>
#include <cstdlib>
#include <ctime>
#include <sys/time.h>
#include <cassert>

#define ARR_SIZE 1000000

using std::string;

void printtime (struct timeval& start, struct timeval& end, string str);   

int main (void)
{
  int arr[ARR_SIZE];
  int tmp;
  struct timeval start, stop;

  srand (time (NULL));

  /* Writing data to array */
  gettimeofday (&start, NULL);
  for (int i = 0; i < ARR_SIZE; i++)
  {
    arr[i] = rand();
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to write to array."));

  /* Reading data from array */
  gettimeofday (&start, NULL);
  for (int i = 0; i < ARR_SIZE; i++)
  {
    tmp = arr[i];
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to read from array sequentially."));

  /* Reading data from array randomly*/
  gettimeofday (&start, NULL);
  for (int i = 0; i < ARR_SIZE; i++)
  {
    tmp = arr[rand() % ARR_SIZE];
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to read from array randomly."));


  int *darr = (int *) calloc (sizeof (int), ARR_SIZE);  

  /* Writing data to array */
  gettimeofday (&start, NULL);
  for (int i = 0; i < ARR_SIZE; i++)
  {
    darr[i] = rand();
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to write to dynamic array."));

  /* Reading data from array */
  gettimeofday (&start, NULL);
  for (int i = 0; i < ARR_SIZE; i++)
  {
    tmp = darr[i];
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to read from dynamic array sequentially."));

  /* Reading data from dynamic array randomly*/
  gettimeofday (&start, NULL);
  for (int i = 0; i < ARR_SIZE; i++)
  {
    tmp = darr[rand() % ARR_SIZE];
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to read from dynamic array randomly."));

  std::vector<int> v(ARR_SIZE);
  assert (v.capacity() == ARR_SIZE);

  /* Writing to vector using indices*/
  gettimeofday (&start, NULL);
  for (int i = 0; i < ARR_SIZE; i++)
  {
    v[i] = rand();
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to write to vector using indices."));
  assert (v.capacity() == ARR_SIZE);

  /* Reading from vector using indices*/
  gettimeofday (&start, NULL);
  for (int i = 0; i < ARR_SIZE; i++)
  {
    tmp = v[i];
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to read from vector using indices, sequentially."));

  /* Reading data from dynamic array randomly*/
  gettimeofday (&start, NULL);
  for (int i = 0; i < ARR_SIZE; i++)
  {
    tmp = v[rand() % ARR_SIZE];
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to read from vector using indices, randomly."));

  std::vector<int> v2(ARR_SIZE);

  /* Writing to vector using iterators*/
  gettimeofday (&start, NULL);
  std::vector<int>::iterator itr, itr_end;
  for (itr = v2.begin(), itr_end = v2.end(); itr != itr_end; itr++)
  {
    *itr = rand();
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to write to vector using iterators."));


  /* Reading from vector using iterators*/
  gettimeofday (&start, NULL);
  for (itr = v2.begin(), itr_end = v2.end(); itr != itr_end; itr++)
  {
    tmp = *itr;
  }
  gettimeofday (&stop, NULL);
  printtime (start, stop, string ("Time taken to read from vector using iterators."));

  return 0;
}

void printtime (struct timeval& start, struct timeval& end, string str)
{
  double start_time, end_time, diff;

  start_time = ((start.tv_sec) * 1000 + start.tv_usec/1000.0);
  end_time   = ((end.tv_sec) * 1000 + end.tv_usec/1000.0);
  diff = end_time - start_time;

  std::cout << str << " : " << diff << " ms" << std::endl;
}

---

编辑

正如评论中所建议的,这里有更多信息: -

• 编译器: - g ++ - 4.5.2
• 标志: - 无(即默认值)
• 优化: - 无(我想在通常的设置中测试行为.优化可能会改变程序的行为,例如,因为从不使用变量tmp,读取矢量/数组的步骤可能会被完全跳过或缩小为只是最后的任务.至少这是我所理解的).

Answer 1

当然不是一个明确的答案，但是您正在循环中写入变量，这意味着编译器可以轻松猜测顺序读取的最终结果应该是什么，从而优化循环。由于它显然没有这样做，我认为没有优化，这绝对不利于迭代器方法。其他数字太接近，无法得出结论。