根据sgi 的STL源码的二级分配算法改写的内存池分配程序，只要稍微修改就可以实现共享内存方式治理，使用C++标准库容器中的map，set，multimap，multiset测试通过，vector测试通不过，原因是在内存回收的时候考虑的比较简单，vector每次分配内存个数不固定，回收也不固定，这样的话，程序还需要继续完善。
内存池治理程序源码如下：xml:namespace PRefix = sizeof(Cookie)
以下是引用片段：
#ifndef MY_ALLOCATOR_H_
#define MY_ALLOCATOR_H_
#include "stdafx.h"
#include <limits>
#include <iostream>
namespace happyever
{
enum { NODENUMS = 2 };
union _Obj
{
union _Obj* M_free_list_link;
char M_client_data[1];
} ;
typedef union _Obj Obj;
strUCt _Cookie
{
int iShmKey; /* 共享内存键值 */
int iShmID; /* iShmKey对应的shmid */
int iSemKey; /* 锁信号键值 */
int iSemID; /* 锁信号标识 */
int iTotalsize; /* 容器总容量 */
void* pStartall; /* 共享内存自身地址 */
char* pStartfree; /* 自由空间的开始地址*/
char* pEndfree; /* 自由空间的结束地址*/
int iUseNum[NODENUMS];
/*用来存放free_list中节点的size*/
short sFreelistIndex[NODENUMS];
/*存放分配内存节点的链表*/
Obj* uFreelist[NODENUMS];
};
typedef struct _Cookie Cookie;
//Obj;
//Cookie;
static Cookie *pHead = NULL;
template <class T>
class MyAlloc
{
private:
static const int ALIGN = sizeof(Obj);
int round_up(int bytes);
int freelist_index(int bytes);
int freelist_getindex(int bytes);
char* chunk_alloc(int size, int *nobjs);
void* refill(int num,int n);
public:
// type definitions
typedef T value_type;
typedef T* pointer;
typedef const T* const_pointer;
typedef T& reference;
typedef const T& const_reference;
typedef std::size_t size_type;
typedef std::ptrdiff_t difference_type;
template <class U>
struct rebind
{
typedef MyAlloc<U> other;
};
pointer address (reference value) const
{
return &value;
}
const_pointer address (const_reference value) const
{
return &value;
}
MyAlloc() throw()
{
std::cout<<"MyAlloc"<<std::endl;
}
MyAlloc(const MyAlloc& x) throw()
{
std::cout<<"const MyAlloc"<<std::endl;
}
template <class U>
MyAlloc (const MyAlloc<U>& x) throw()
{
std::cout<<"const MyAlloc<U>"<<std::endl;
}
~MyAlloc() throw()
{
std::cout<<"~MyAlloc"<<std::endl;
}
size_type max_size () const throw()
{
return std::numeric_limits<std::size_t>::max() / sizeof(T);
}
//void PrintFreelistAndCookie();
pointer allocate (size_type num, const void* = 0)
{
pointer ret = 0;
Obj** my_free_list;
Obj* result;
int index;
// print message and allocate memory with global new
std::cerr << "allocate " << num << " element(s)"
<< " of size " << sizeof(T) << std::endl;
index = freelist_index(sizeof(T));
if(index >= NODENUMS)
{
return NULL;
}
my_free_list = pHead->uFreelist + index;
//Lock(semid,LOCK_NUM);
result = *my_free_list;
if (result == 0)
{
ret = (pointer)refill((int)num, round_up(sizeof(T)));
}
else
{
*my_free_list = result->M_free_list_link;
ret = (pointer)result;
}
//UnLock(semid,LOCK_NUM);
pHead->iUseNum[index] = pHead->iUseNum[index] + (int)num;
if(0 == ret)
{
std::cerr << "alloc memory fail!" << std::endl;
exit(1);
}
std::cerr << " allocated at: " << (void*)ret << std::endl;
PrintFreelistAndCookie();
return ret;
}
void construct (pointer p, const T& value)
{
// initialize memory with placement new
new((void*)p)T(value);
}
void destroy (pointer p)
{
// destroy objects by calling their destructor
p->~T();
}
void deallocate (pointer p, size_type num)
{
Obj** my_free_list;
Obj* q ;
int index;
index = freelist_getindex(sizeof(T));
if(index >= NODENUMS)
{
std::cerr << "deallocate memory fail!" << std::endl;
exit(1);
}
my_free_list = pHead->uFreelist + index;
q = (Obj*) p;
//Lock(semid,LOCK_NUM);
/*这个地方可能会有问题*/
//for(int i=0 ;i<(int)num ; i++)
{
q->M_free_list_link = *my_free_list;
*my_free_list = q;
}
//UnLock(semid,LOCK_NUM);
pHead->iUseNum[index] = pHead->iUseNum[index] - (int)num;
std::cerr << "deallocate " << num << " element(s)"
<< " of size " << sizeof(T)
<< " at: " << (void*)p << std::endl;
PrintFreelistAndCookie();
}
};
template <class T>
int MyAlloc<T>::round_up(int bytes)
{
int i;
i = bytes;
if(bytes < ALIGN)
{
i = ALIGN;
}
std::cout<<"round_up:bytes="<<bytes<<" , return="<<i<<std::endl;
return i;
};
template <class T>
int MyAlloc<T>::freelist_index(int bytes)
{
int i;
for(i=0 ; i< NODENUMS ; i++)
{
if(pHead->sFreelistIndex[i] == bytes)
break;
}
if(i >= NODENUMS)
{
for(i=0 ; i< NODENUMS ; i++)
{
if(pHead->sFreelistIndex[i] == 0)
{
pHead->sFreelistIndex[i] = bytes;
std::cout<<"freelist_index:bytes="<<bytes<<" , return="<<i<<std::endl;
return i;
}
}
}
std::cout<<"freelist_index:bytes="<<bytes<<" , return="<<i<<std::endl;
return i;
};
template <class T>
int MyAlloc<T>::freelist_getindex(int bytes)
{
int i;
for(i=0 ; i< NODENUMS ; i++)
{
if(pHead->sFreelistIndex[i] == bytes)
break;
}
std::cout<<"freelist_getindex:bytes="<<bytes<<" , return="<<i<<std::endl;
return i;
};
template <class T>
char* MyAlloc<T>::chunk_alloc(int size, int *nobjs)
{
char* result;
int counts = *nobjs;
int total_bytes = size * counts;
int bytes_left = int(pHead->pEndfree - pHead->pStartfree);
std::cout<<"chunk_alloc:total_bytes = "<<total_bytes
<<",bytes_left = "<<bytes_left<<std::endl;
if (bytes_left >= total_bytes)
{
result = pHead->pStartfree;
pHead->pStartfree += total_bytes;
std::cout<<"chunk_alloc:total_bytes = "<<total_bytes
<<",result = "<<*result<<",start_free = "<<&(pHead->pStartfree)<<std::endl;
}
else if (bytes_left >= size)
{
counts = bytes_left/size;
total_bytes = size * counts;
result = pHead->pStartfree;
pHead->pStartfree += total_bytes;
*nobjs = counts;
std::cout<<"chunk_alloc:total_bytes = "<<total_bytes<<",nobjs = "<<nobjs
<<",result = "<<*result<<",start_free = "<<&(pHead->pStartfree)<<std::endl;
}
else
{
/*还需要处理回收其他空闲freelist里面的空间*/
result = NULL;
}
return(result);
};
template <class T>
void* MyAlloc<T>::refill(int num,int n)
{
int counts = num;
int *nobjs = &counts;
char* chunk;
Obj** my_free_list;
Obj* result;
Obj* current_obj;
Obj* next_obj;
int i;
chunk = chunk_alloc(n, nobjs);
if(chunk == NULL)
{
return(chunk);
}
counts = *nobjs;
if (1 == counts)
{
return(chunk);
}
my_free_list = pHead->uFreelist + freelist_index(n);
result = (Obj*)chunk;
*my_free_list = next_obj = (Obj*)(chunk + n*num);
for (i = 1; ; i++)
{
current_obj = next_obj;
next_obj = (Obj*)((char*)next_obj + n);
if (counts - 1 == i)
{
current_obj->M_free_list_link = 0;
break;
}
else
{
current_obj->M_free_list_link = next_obj;
}
}
return(result);
};
/*这个函数可以改写成自己的共享内存分配函数*/
static void InitShm()
{
int i,size=1000;
pHead = (Cookie*)malloc(sizeof(Cookie)+size);
pHead->iTotalsize = sizeof(Cookie)+size;
pHead->pStartall = pHead;
pHead->pStartfree = (char*)pHead + sizeof(Cookie);
pHead->pEndfree = (char*)pHead + pHead->iTotalsize;
for(i=0 ; i <NODENUMS ; i++)
{
pHead->sFreelistIndex[i]=0;
pHead->uFreelist[i]=0;
pHead->iUseNum[i]=0;
}
}
static void PrintFreelistAndCookie()
{
int i,j;
Obj* my_free_list;
std::cout<<"Cookie info :"<<std::endl;
std::cout<<"sizeof(struct Cookie) = "<<sizeof(Cookie)<<std::endl;
std::cout<<"Totalsize = "<<pHead->iTotalsize<<std::endl;
std::cout<<"UsedSize = "<<int(pHead->pStartfree-(char*)pHead)<<std::endl;
std::cout<<"FreepoolSize = "<<int(pHead->pEndfree - pHead->pStartfree)<<std::endl;
std::cout<<"Startall = "<<&(pHead->pStartall)<<std::endl;
std::cout<<"Startfree = "<<&(pHead->pStartfree)<<std::endl;
std::cout<<"Endfree = "<<&(pHead->pEndfree)<<std::endl;
std::cout<<"nFreelist info :"<<std::endl;
for(i=0 ; i<NODENUMS ; i++)
{
j=0;
std::cout<<"iUseNum["<<i<<"] = "<<pHead->iUseNum[i]<<std::endl;
std::cout<<"FreelistIndex["<<i<<"] = "<<pHead->sFreelistIndex[i]<<std::endl;
my_free_list = pHead->uFreelist[i];
if(my_free_list->M_client_data != 0)
{
while(my_free_list->M_client_data != 0)
{
j++;
my_free_list = my_free_list->M_free_list_link;
}
std::cout<<"free_list["<<i<<"]; node counts="<<j<<std::endl;
}
}
}
template <class T1, class T2>
bool Operator== (const MyAlloc<T1>&,const MyAlloc<T2>&) throw()
{
return true;
}
template <class T1, class T2>
bool operator!= (const MyAlloc<T1>&,const MyAlloc<T2>&) throw()
{
return false;
}
}
#endif /*MY_ALLOCATOR_H_*/
测试程序的源码如下：
// MyStl.cpp : 定义控制台应用程序的入口点。
//
#include "stdafx.h"
#include <map>
#include <vector>
#include <string>
#include <utility>
#include <iostream>
#include "MyAlloc.h"
using namespace std;
int _tmain(int argc, _TCHAR* argv[])
{
happyever ::InitShm();
multimap<string,int,less<string>,happyever ::MyAlloc<string> > m;
m.insert(make_pair(string("Harry"), 32));
m.insert(make_pair(string("Mary"), 59));
m.insert(make_pair(string("Roger"), 18));
m.insert(make_pair(string("Nancy"), 37));
m.insert(make_pair(string("Mary"), 23));
typedef multimap<string,int,less<string>,happyever ::MyAlloc<string> >::iterator Iter;
for (Iter p = m.begin(); p != m.end(); p++)
{
cout << p->first << "," << p->second << endl;
}
Iter p = m.find("Harry");
m.erase(p);
/*p = m.find("Harry");
cout << "Harry is: " << p->second << "." << endl;*/
for (Iter p = m.begin(); p != m.end(); p++)
{
cout << p->first << "," << p->second << endl;
}
return 0;
}
以上程序在vs2005，vc6上测试通过。使用MinGW编译的时候只需要去掉vc的预编译头文件
#include "stdafx.h"
即可。
以上程序只要稍微修改，就可以实现共享内存的治理，可以方便的使用标准库提供的容器。加上信号量的锁机制。
以上为了学习而改写的SGI的stl二级分配算法实现的。以上代码存在一定的局限性。我另外完整实现了共享内存治理的STL标准的alloctor程序，使用posix信号量加锁。目前应用在aix的xlC编译环境下。因为源码涉及公司的商业秘密，所以不能公开。但基本上以上源码已经体现了自己治理内存的完整思路，供这方面需求的朋友一起学习研究用。
