Re: Vectors vs Arrays performance

From:

ytrembla@nyx.nyx.net (Yannick Tremblay)

Newsgroups:

comp.lang.c++

Date:

05 Jan 2010 12:38:38 GMT

Message-ID:

<1262695118.550420@irys.nyx.net>

In article <EtKdnc1YleZWC93WnZ2dnUVZ8qqdnZ2d@giganews.com>,
Leigh Johnston <leigh@i42.co.uk> wrote:

We can come up with examples and counter-examples all day long.

Indeed, so let's have a little modification to only consider operator[] since that is what your original statement referred to

Consider this:

int main()
{
    std::vector<int> v;
    v.reserve(1000);
    lib::vecarray<int, 1000> va;
    {
        timer t("vector: ");
        for (int i = 0; i != 1000000; ++i)
        {
            for (int j = 0; j != 1000; ++j)
                v.push_back(j);
            v.clear();
        }
    }
    {
        timer t("vecarray: ");
        for (int i = 0; i != 1000000; ++i)
        {
            for (int j = 0; j != 1000; ++j)
                va.push_back(j);
            va.clear();
        }
    }
}

So let's make a few mods to prefill the vectors since the claim
referred to operator[] not push_back() nor clear()

#include <vector>
#include "vecarray.h"
#include <boost/progress.hpp>
#include <iostream>

size_t const VSIZE = 100000;
size_t const LOOP = 100000;

int main()
{
  lib::vecarray<int, VSIZE> va;
  va.resize(VSIZE);
  std::vector<int> v;
  v.resize(VSIZE);
  {
    std::cout << "vector: ";
    boost::progress_timer t;
    for (size_t i = 0; i < LOOP; ++i)
      {
    for (size_t j = 0; j < v.size(); ++j)
      v[j] = j;
      }
  }
  {
    std::cout << "vecarray: ";
    boost::progress_timer t;
    for (size_t i = 0; i < LOOP; ++i)
      {
    for (size_t j = 0; j < va.size() ; ++j)
      va[j] = j;
      }
  }
}

Compiled with g++ -std=gnu++0x -O3

vector: 4.70 s
vecarray: 5.72 s

Over a few runs. The vector varies from 4.69s to 4.71s but the
vecarray varies greatly from 5.14s to 6.09s

Hmm, interesting.... Care to explain what's happening and why is the
vecarray now slower than std::vector if the statement the operator[]
for a vector will always be slower than for a vecarray is true?

I am sure this benchmark has flaws but it is not obviously more flawed
than the previously posted ones. Neither are particularly realistic
but they show opposite and one of then shows what is arguably against
intuition...

Which outputs (several runs done, VC++ release build, secure STL disabled):

vector: 2.9741 seconds
vecarray: 2.6782 seconds

Which indicates (for this example) that a stack based container
(http://i42.co.uk/stuff/vecarray.htm) is 11% quicker than std::vector.

1) std::vector contains a pointer to its elements, a stack based container
(or an ordinary array on the stack) does not
2) you may benefit from the locality of your container and other local
variables (think CPU cache)

Have you seen the film "The Matrix"? "There is no spoon." "There is no
pointer.".

So how would The Matrix explain the faster std::vector then? Is there
a fork somewhere?

Regards

Yannick