Search code examples
c++arraysperformancebooststl

Very fast object allocator for small object of same size

I just have to write some code that has to have best performance.

Requirements:

I need a very fast object allocator for quick creation of object. My object as only 3 doubles in it. Allocation and deallocation will occurs one object at a time only.

I made a lots of research and come up with:

std::vector<MyClass, boost::fast_pool_allocator<MyClass>>

I wonder (in 2014-07):

  • Does stl have something equivalent to boost::boost:fast_pool_allocator ?
  • Is there a better solution to what I have found ?

There is additional information to answer some comments:

  • The code will be used to optimize my algorithm for: Code Project article on Convex Hull
  • I need to convert C# code to C or C++ to improve performance. I should compete against another algorithm written in pure "C". I just discover that my comparison chart in my article have errors because I tested against code compiled in C for x86-Debug. In x64-release the "C" code is a lot faster (a factor of 4 to 5 times faster than in x86-debug).
  • According to This Boost documentation and this Answer at StackOverFlow, boost:fast_pool_allocator seems to be the best allocator to use for small memory chunk of same size query one by one. But I would like to make sure nothing else exists that is either more standard (part of stl) or faster.
  • My code will be developed on Visual Studio 2013 and target any windows platform (no phone or tablet).
  • My intend is not to have fast code, it is to have the fastest code. I prefer not having too much twisted code if possible and also look for code that is maintainable (at least a minimum).
  • If possible, I also would like to know the impact of using std:vector vs array (ie: []).
  • For more info, you could see Wikipedia - Object pool pattern
Solution

  • The closest thing to what I was looking for was Paulo Zemek Code Project article: O(1) Object Pool in C++.

    But I finally did allocate/reserve memory of size= the maximum size possible * my object size. Because I was not using any objects that required to live longer than my algorithm loop, I cheated by saying that location in reserved memory space was object. After my algorithm loop, I flushed the reserved memory space. It appears to me to be the fastest. Very not elegant but extremely fast and only require one allocation and one deallocation.

    I was not totally satisfy with the answer, that's why I answered myself. I also added a comments to the questions about every comments and added this answer to make thinks clear. I know that my decision/implementation was not totally in accordance with the question but I think that it should have conducted to something similar.