memset on vector

Question

As per Mark Ransoms answer on using memset, i am using memset on a vector<int> to assign values to all the elements.

memset(&match_begin[0], 0xff , sizeof(match_begin[0]) * match_begin.size());

It does have a significant performance improvement over std::fill and it works fine(g++ 4.3.2, 64bit linux). Is this code safe , as in , would std::vector implementations always guarantee that memory allocation for the data will be contiguous? Is it possible that in a future(or different) implementation of the stl library that this might change and break my code later?

`std::fill` is clearly cleaner, and the only reason to consider `memset` is performance. How did you measure the performance difference? gcc has an optimization that recognizes memset-like loops and replaces them with a call to memset. — Marc Glisse, Apr 08 '13 at 11:52
@Marc, in my function i had accidentally put both memset and fill. Actually there is not any visibile difference in performance. My code reads through each packet in a tcp stream. I get a basic idea of performance by sending a large file and measuring time. — woodstok, Apr 08 '13 at 12:30

ForEveR · Accepted Answer · 2013-04-08T09:56:36.110

5

would std::vector implementations always guarantee that memory allocation for the data will be contiguous

Yes. 23.3.6.1/1. In C++03 standard there are equal strings at 23.2.4/1

The elements of a vector are stored contiguously, meaning that if v is a vector where T is some type other than bool, then it obeys the identity &v[n] == &v[0] + n for all 0 <= n < v.size()

Is it possible that in a future(or different) implementation of the stl library that this might change and break my code later?

No. Vector should be contiguous always.

However, in gcc 4.6.3 there is only one optimization of fill, using memset. This optimization is for char types

  // Specialization: for char types we can use memset.
  template<typename _Tp>
    inline typename
    __gnu_cxx::__enable_if<__is_byte<_Tp>::__value, void>::__type
    __fill_a(_Tp* __first, _Tp* __last, const _Tp& __c)
    {
      const _Tp __tmp = __c;
      __builtin_memset(__first, static_cast<unsigned char>(__tmp),
               __last - __first);
    }

edited Apr 08 '13 at 09:56

answered Apr 08 '13 at 09:47

ForEveR

55,233
2
119
133

Where can i find any documentation that validates this? – woodstok Apr 08 '13 at 09:49
even though I have the feeling that you're right, can you still provide some source ? – lucasg Apr 08 '13 at 09:50
Thank you for the section numbers, but where have you pulled this info? Can you provide a link? – woodstok Apr 08 '13 at 09:54
2

@MIkhail I just put the link there. – Michael Wild Apr 08 '13 at 09:56

score 1 · Answer 2 · answered Apr 08 '13 at 09:50

1

This should be safe from a memory perspective.

Just bear in mind that if the content of your vector is not a plain data type, you shouldn't be doing this sort of thing.

answered Apr 08 '13 at 09:50

Mats Petersson

126,704
14
140
227

Yes, Thats true. it has to be POD. – woodstok Apr 08 '13 at 09:51

score 1 · Answer 3 · answered Apr 08 '13 at 09:57

memset fails with vector of bool type, just try below example, though as other mentioned it work for other plain data type.I shared it just to make it explicit its not true for all plain data types.

#include <vector>
using namespace std;
int _tmain(int argc, _TCHAR* argv[])
{
    vector<bool>vec;
    vec.push_back(0);
    vec.push_back(1);
    memset(&vec[0], 0, sizeof(vec[0]) * vec.size());
    return 0;
}

memset on vector

3 Answers3