最近写的一个程序中需要频繁的申请和释放零碎的内存块，这些内存块的大小却只有简单的几种。如果直接使用系统调用malloc/free、new/delete进行内存分配和释放，则效率很低。程序运行时间长了会产生大量的内存碎片。想起uC/OS-II 里面提供了个内存分配单元，正好满足我的需要。就把里面相关的代码扒了出来。写成了一个内存池的类。

这个内存池的功能非常的简单，初始化时分配一大块内存，然后将各个小内存块串成一个单项链表。每次分配内存块时从链表的头上去取一个内存块。回收内存块时也是将内存块插到链表的开头处。

这个类的结构如下：

#ifndef MEMPOOL_H
#define MEMPOOL_H

#define MEM_NO_ERR          0
#define MEM_INVALID_ADDR    1
#define MEM_INVALID_BLKS    2
#define MEM_INVALID_SIZE    3
#define MEM_INVALID_PART    4
#define MEM_INVALID_PBLK    5
#define MEM_FULL            6
class MemPool
{
public:
	MemPool();
	~MemPool();
    int create (int nblocks, unsigned int blocksize);
    void* get( void );
    int release ( void *pblk );
    int blocks( void ) const {return m_memNBlks;};
    int frees( void ) const {return m_memNFree;};
private:
    char *m_memAddr; /* Pointer to beginning of memory partition */
    char *m_memFreeList; /* Pointer to list of free memory blocks */
    int m_memBlkSize; /* Size (in bytes) of each block of memory */
    int m_memNBlks; /* Total number of blocks in this partition */
    int m_memNFree; /* Number of memory blocks remaining in this */
};

#endif

create 函数初始化内存池。主要的工作就是分配内存，然后将内存块串起来形成一个链表。因为要用指针形成链表，因此要求内存块的大小至少要能容纳一个指针。

get 函数获得一个内存块，如果没有剩余的内存块了，就返回 null。

release 函数回收内存块。

blocks 函数返回内存池总共有多少内存块。

frees 函数返回内存池还剩多少剩余的内存块。

代码实现如下：

#include <stddef.h>
#include "MemPool.h"

MemPool::MemPool()
{
    m_memAddr = NULL;
    m_memFreeList = NULL;
    m_memBlkSize = 0;
    m_memNBlks = 0;
    m_memNFree = 0;
}
MemPool::~MemPool()
{
    if(m_memAddr != NULL)
    {
        delete [] m_memAddr;
    }
}
int MemPool::create ( int nblks, unsigned int blksize )
{
    if( m_memAddr != NULL )
    {
        delete [] m_memAddr;
    }
    m_memAddr = new char[nblks * blksize];
    if ( m_memAddr == NULL )
    {
        return MEM_INVALID_ADDR;
    }
    if ( nblks < 2 )
    {
        /* Must have at least 2 blocks per partition */
        return MEM_INVALID_BLKS;
    }

    if ( blksize < sizeof(void *) )
    {
        /* Must contain space for at least a pointer */
        return MEM_INVALID_SIZE;
    }
    void ** p = (void **)m_memAddr;
    char *pblk = m_memAddr + blksize;
    for (int i = 0; i < (nblks - 1); i++)
    {
        *p = (void *) pblk;
        p = (void **) pblk;
        pblk   = pblk + blksize;
    }
    *p = (void *)0;

    m_memFreeList = m_memAddr;
    m_memNBlks = nblks;
    m_memNFree = nblks;
    return MEM_NO_ERR;
}

void * MemPool::get( void )
{
    void *pblk;
    if (m_memNFree > 0)
    {
        /* See if there are any free memory blocks */
        pblk = m_memFreeList;    /* Yes, point to next free memory block */
        m_memFreeList = (char *) *(void **)pblk; /* Adjust pointer to new free list */
        m_memNFree--;/* One less memory block in this partition  */
        return (pblk); /* Return memory block to caller */
    }
    return ((void *)0);
}
int MemPool::release ( void *pblk )
{
    if (pblk == (void *)0)
    {
        /* Must release a valid block */
        return (MEM_INVALID_PBLK);
    }
    if (m_memNFree >= m_memNBlks)
    {
        /* Make sure all blocks not already returned */
        return (MEM_FULL);
    }
    /* Insert released block into free block list */
    *(void **)pblk = m_memFreeList;
    m_memFreeList = (char *) pblk;

    m_memNFree++; /* One more memory block in this partition */
    return (MEM_NO_ERR); /* Notify caller that memory block was released */
}

#include <iostream>
#include "MemPool.h"
using namespace std;
typedef struct
{
    unsigned char *ImageData;
    int SizeX;
    int SizeY;
    int ImageID;
    double Timestamp;
    double TransferTime;
    unsigned int PacketCount;
}IMAGE_INFO;

int main()
{
    IMAGE_INFO * p[15];
    MemPool mem;
    mem.create(15, sizeof(IMAGE_INFO));

    for(int i = 0; i < 10; i++)
    {
        p[i] = (IMAGE_INFO *)mem.get();
        cout << "p[" << i << "] addr = " << p[i] << endl;
        p[i]->SizeX = i;
        p[i]->SizeY = i;
    }
    for(int i = 1; i < 10; i++)
    {
        cout << "p[" << i << "]->SizeX = " << p[i]->SizeX << endl;
        mem.release(p[i]);
    }
    cout << "mem.blocks()" << mem.blocks() << endl;
    cout << "mem.frees()" << mem.frees() << endl;
    return 0;
}

这个测试程序的运行结果如下：

p[0] addr = 0xac1358
p[1] addr = 0xac1380
p[2] addr = 0xac13a8
p[3] addr = 0xac13d0
p[4] addr = 0xac13f8
p[5] addr = 0xac1420
p[6] addr = 0xac1448
p[7] addr = 0xac1470
p[8] addr = 0xac1498
p[9] addr = 0xac14c0
p[1]->SizeX = 1
p[2]->SizeX = 2
p[3]->SizeX = 3
p[4]->SizeX = 4
p[5]->SizeX = 5
p[6]->SizeX = 6
p[7]->SizeX = 7
p[8]->SizeX = 8
p[9]->SizeX = 9
mem.blocks()15
mem.frees()14

最后多说一句，如果程序中多个线程要访问同一个内存池，那个需要给 get 和 release 函数加锁。

另外，这个代码其实可以用C++的模版类来实现。等我有空了就写一个。

一个及其简化版本的内存池实现