C++中一个高效的内存池实现

原文出处：http://www.cppblog.com/weiym/archive/2012/05/05/173785.html

在高效C++编程中看到一个不错的内存池实现方案，这里共享下，大家看看有什么不足。
代码很简单，如下：
template<typename T>
class CMemoryPool
{
    public:
        enum { EXPANSION_SIZE = 32};

        CMemoryPool(unsigned int nItemCount = EXPANSION_SIZE)
        {
            ExpandFreeList(nItemCount);
        }
       
        ~CMemoryPool()
        {
            //free all memory in the list
            CMemoryPool<T>* pNext = NULL;
            for(pNext = m_pFreeList; pNext != NULL; pNext = m_pFreeList)
            {
                m_pFreeList = m_pFreeList->m_pFreeList;
                delete [](char*)pNext;
            }
        }

        void* Alloc(unsigned int /*size*/)
        {
            if(m_pFreeList == NULL)
            {
                ExpandFreeList();
            }
           
            //get free memory from head
            CMemoryPool<T>* pHead = m_pFreeList;
            m_pFreeList = m_pFreeList->m_pFreeList;
            return pHead;
        }

        void Free(void* p)
        {
            //push the free memory back to list
            CMemoryPool<T>* pHead = static_cast<CMemoryPool<T>*>(p);
            pHead->m_pFreeList = m_pFreeList;
            m_pFreeList = pHead;
        }

    protected:
        //allocate memory and push to the list
        void ExpandFreeList(unsigned nItemCount = EXPANSION_SIZE)
        {
            unsigned int nSize = sizeof(T) > sizeof(CMemoryPool<T>*) ? sizeof(T) : sizeof(CMemoryPool<T>*);
            CMemoryPool<T>* pLastItem = static_cast<CMemoryPool<T>*>(static_cast<void*>(new char[nSize]));
            m_pFreeList = pLastItem;
            for(int i=0; i<nItemCount-1; ++i)
            {
                pLastItem->m_pFreeList = static_cast<CMemoryPool<T>*>(static_cast<void*>(new char[nSize]));
                pLastItem = pLastItem->m_pFreeList;
            }

            pLastItem->m_pFreeList = NULL;
        }

    private:
        CMemoryPool<T>* m_pFreeList;
};

它的实现思想就是每次从List的头上取内存， 如果取不到则重新分配一定数量； 用完后把内存放回List头部，这样的话效率很高，因为每次List上可以取到的话，肯定是空闲的内存。

当然上面的代码只是针对单线程的，要支持多线程的话也很简单，外面加一层就可以了，
代码如下：
class CCriticalSection
{
public:
    CCriticalSection()
    {
        InitializeCriticalSection(&m_cs);
    }

    ~CCriticalSection()
    {
        DeleteCriticalSection(&m_cs);
    }

    void Lock()
    {
        EnterCriticalSection(&m_cs);
    }

    void Unlock()
    {
        LeaveCriticalSection(&m_cs);
    }

protected:
    CRITICAL_SECTION m_cs;
};

template<typename POOLTYPE, typename LOCKTYPE>
class CMTMemoryPool
{
    public:
        void* Alloc(unsigned int size)
        {
            void* p = NULL;
            m_lock.Lock();
            p = m_pool.Alloc(size);
            m_lock.Unlock();

            return p;
        }

        void Free(void* p)
        {
            m_lock.Lock();
            m_pool.Free(p);
            m_lock.Unlock();   
        }

    private:
        POOLTYPE m_pool;
        LOCKTYPE m_lock;
};

这是我的测试代码：
#include <iostream>
#include <windows.h>

using namespace std;

#include "MemoryPool.h"
#include "MTMemoryPool.h"

class CTest
{
public:
    int m_n;
    int m_n1;

    void* operator new(size_t size)
    {
        void* p = s_pool->Alloc(size);
        return p;
    }

    void operator delete(void* p, size_t size)
    {
        s_pool->Free(p);
    }

    static void NewPool()
    {
        //s_pool = new CMemoryPool<CTest>;
        s_pool = new CMTMemoryPool<CMemoryPool<CTest>, CCriticalSection>;
    }

    static void DeletePool()
    {
        delete s_pool;
        s_pool = NULL;
    }
   
    //static CMemoryPool<CTest>* s_pool;
    static CMTMemoryPool<CMemoryPool<CTest>, CCriticalSection>* s_pool;
};

//CMemoryPool<CTest>* CTest::s_pool = NULL;
CMTMemoryPool<CMemoryPool<CTest>, CCriticalSection>* CTest::s_pool = NULL;

void testFun()
{
    int i;
    const int nLoop = 10;
    const int nCount = 10000;
   
    for(int j = 0; j<nLoop; ++j)
    {
        typedef CTest* LPTest;
        LPTest arData[nCount];
        for(i=0;i <nCount; ++i)
        {
            arData[i] = new CTest;
        }

        for(i=0;i <nCount; ++i)
        {
            delete arData[i];
        }
    }
}

int main(int argc, char* argv[])
{
    {
        unsigned int dwStartTickCount = GetTickCount();

        CTest::NewPool();

        testFun();
       
        CTest::DeletePool();
       
        cout << "total cost" << GetTickCount() - dwStartTickCount << endl;
    }

    system("pause");

    return 0;
}
在我机器上测试结果比系统默认的CRT实现高效N倍。