2、多线程内存池

 

上一节很简略的说了下单线程内存池,单线程内存池如果要放在多线程环境下使用是不安全的,我们需要进行保护,如何保护,最简单的方法就是加临界区,云风的实现里面是用原子操作模拟一个临界区,我实测跟临界区性能非常接近,甚至很多时候不如临界区,所以我就不追求用原子操作模拟临界区了,还是直接用临界区最简单。

 

class CMemPool

{

public:

        struct memory_list

        {

                memory_list *_next;

        };

        struct alloc_node

        {

                size_t _size;

                size_t _number;

                size_t _bksize;

                long _guard;

                memory_list *_free_list;

        };

        struct chunk_list

        {

                chunk_list *_next;

                memory_list *_data;

                size_t _size;

                size_t _idx;

        };

 

        ~CMemPool();

        static CMemPool &instance()

        {

                if(!_instance)

                {

                        create_instance();

                }

                return *_instance;

        }

        static int chunk_index(size_t size);

        void *allocate(size_t size, size_t *psize=NULL);

        void deallocate(void *p, size_t size);

 

        //以下为几个检测和设置预分配参数的函数,2007.06.08

        size_t getallocnumber(size_t size, size_t *psize=NULL);

        size_t setallocnumber(size_t size, size_t number);

        static void dumpallocnode();

 

private:

        static alloc_node _vnode[92];

 

        chunk_list *_chunk_list;

        long _chunk_guard;

 

        static CMemPool *_instance;

        static long _singleton_guard;

        static bool _singleton_destroyed;

        static void create_instance();

        static void build_chunknode();

 

        CMemPool();

        memory_list *alloc_chunk(size_t idx);

};

 

 

CMemPool *CMemPool::_instance = 0;

long CMemPool::_singleton_guard = 0;

bool CMemPool::_singleton_destroyed = false;

CMemPool::alloc_node CMemPool::_vnode[92];

 

struct chunk_desc

{

        int s;                // 区间起始字节

        int e;                // 区间终止字节

        int align; // 区间内分段对齐值

        int number;     // 区间内分段个数,计算属性

}_cd[] =

{

        {0, 1024, 16, 0},

        {1024, 8192, 256, 0}

};

 

 

void CMemPool::create_instance()

{

        thread_guard guard(&_singleton_guard);

        if(_instance)

                return;

        assert(!_singleton_destroyed);

        static CMemPool obj;

        _instance = &obj;

}

 

void CMemPool::build_chunknode()

{

        const int cdnum = sizeof(_cd)/sizeof(_cd[0]);

        int index=0, number;

        for(int j=0; j<cdnum; j++)

        {

                _cd[j].number = (_cd[j].e-_cd[j].s)/_cd[j].align;

                for(int i=0; i<_cd[j].number; i++)

                {

                        int unitsize = (i+1)*_cd[j].align+_cd[j].s;

                        _vnode[index]._size = unitsize;

 

                        if(unitsize < 512)

                                number = 4096/unitsize;

                        else if(unitsize < 1024)

                                number = 16384/unitsize;

                        else

                                number = 65536/unitsize;

                        _vnode[index]._number = number;

                        _vnode[index]._bksize = unitsize * number;

                        _vnode[index]._guard = 0;

                        _vnode[index]._free_list = NULL;

                        ++index;

                }

        }

}

 

size_t CMemPool::getallocnumber(size_t size, size_t *psize/*=NULL*/)

{

        int idx = chunk_index(size);

        if(idx >= 0)

                return _vnode[idx]._number;

        return 0;

}

 

size_t CMemPool::setallocnumber(size_t size, size_t number)

{

        int idx = chunk_index(size);

        if(idx >= 0)

        {

                size_t on = _vnode[idx]._number;

                _vnode[idx]._number = number;

                return on;

        }

        return 0;

}

 

void CMemPool::dumpallocnode()

{

        int size = sizeof(_vnode)/sizeof(_vnode[0]);

        printf("_vnode.size = %d\r\n", size);

        for(int i=0; i<size; ++i)

        {

                printf("vnode.size %d, vnode.number %d\r\n", _vnode[i]._size, _vnode[i]._number);

        }

}

 

int CMemPool::chunk_index(size_t bytes)

{

#if(0)

        int idx = 0;

        const int cdnum = sizeof(_cd)/sizeof(_cd[0]);

 

        if(bytes > _cd[cdnum-1].e)

                idx = -1;

        else

        {

                for(int i=0; i<cdnum; i++)

                {

                        if((bytes > _cd[i].s) && (bytes <= _cd[i].e))

                        {

                                idx += (bytes-_cd[i].s+_cd[i].align-1)/_cd[i].align-1;

                                break;

                        }

                        idx += _cd[i].number;

                }

        }

//     printf("bytes %d idx = %d\r\n", bytes, idx);

        return idx;

#else

        //下面的代码是根据静态数据和上面的代码做了优化的,

        //如果修改了基础数据需要对应的修改下面的代码

        if(bytes > 8192)

        {

                return -1;

        }

        else if(bytes > 1024)

        //     idx = _cd[0].number+(bytes-_cd[1].s+_cd[1].align-1)/_cd[1].align-1;

                return 64+(bytes-1024+255)/256-1;

        //     return _cd[0].number+(bytes-1024+255)/256-1;

        else

        //     idx = (bytes-_cd[0].s+_cd[0].align-1)/_cd[0].align-1;

                return (bytes+15)/16-1;

#endif

}

 

CMemPool::CMemPool()

{

        _chunk_list = NULL;

        _chunk_guard = 0;

        build_chunknode();

}

 

CMemPool::~CMemPool()

{

        int s = 0;

        chunk_list *temp = _chunk_list;

        while(temp)

        {

                ++s;

                temp = temp->_next;

        }

        void **chunk = reinterpret_cast<void **>(malloc(s * sizeof(void *)));

        temp = _chunk_list;

        int i=0;

        while(temp)

        {

                chunk[i] = temp->_data;

                ++i;

                temp = temp->_next;

        }

        for(i=0; i<s; i++)

        {

                free(chunk[i]);

        }

        free(chunk);

 

        _singleton_destroyed = true;

        _instance = 0;

}

 

CMemPool::memory_list *CMemPool::alloc_chunk(size_t idx)

{

        thread_guard guard(&_chunk_guard);

        memory_list *&current_list = _vnode[idx]._free_list;

        if(current_list)

                return current_list;

        const size_t node_size = _vnode[idx]._size;

        const size_t number     = _vnode[idx]._number;

        const size_t chunk_size = node_size * number;

        memory_list *ret = current_list = reinterpret_cast<memory_list *>(malloc(chunk_size));

        memory_list *iter = ret;

        //for(size_t i=0; i<=chunk_size-node_size*2; i+=node_size)

        for(size_t i=0; i<number-1; ++i)

        {

                iter = iter->_next = iter+node_size/sizeof(*iter);

        }

        iter->_next = 0;

 

        return ret;

}

 

void *CMemPool::allocate(size_t size, size_t *psize/*=NULL*/)

{

        int idx = chunk_index(size);

        if(idx < 0)

        {

                if(psize) *psize = size;

                return malloc(size);

        }

        if(psize)

                *psize = _vnode[idx]._size;

 

        thread_guard guard(&_vnode[idx]._guard);

        memory_list *&temp = _vnode[idx]._free_list;

 

        if(!temp)

        {

                memory_list *new_chunk = alloc_chunk(idx);

 

                chunk_list *chunk_node;

                if(chunk_index(sizeof(chunk_list))==idx)

                {

                        chunk_node = reinterpret_cast<chunk_list *>(temp);

                        temp = temp->_next;

                }

                else

                {

                        chunk_node = reinterpret_cast<chunk_list *>(allocate(sizeof(chunk_list)));

                }

 

                thread_guard guard(&_chunk_guard);

                chunk_node->_next = _chunk_list;

                chunk_node->_data = new_chunk;

                chunk_node->_size = _vnode[idx]._bksize;

                chunk_node->_idx = idx;

                _chunk_list = chunk_node;

        }

        void *ret = temp;

        temp = temp->_next;

 

        return ret;

}

 

void CMemPool::deallocate(void *p, size_t size)

{

        int idx = chunk_index(size);

        if(idx < 0)

        {

                free(p);

        }

        else

        {

                memory_list *free_block = reinterpret_cast<memory_list *>(p);

                thread_guard guard(&_vnode[idx]._guard);

               

                memory_list *&temp = _vnode[idx]._free_list;     //_free_list[idx];

                free_block->_next = temp;

                temp = free_block;

        }

}

 

以上基本是云风内存池的一个简单修改版,这种模式的内存池由于每次分配释放都要lock unlock,所以效率很低,大概只相当于malloc/free 2-4倍的速度,提速也不是很明显,大概相当于nedmalloc速度的一半左右。

发表于 @ 2010年02月04日 16:54:00 | 评论( 0 ) | 编辑| 举报| 收藏

旧一篇:单线程内存池 | 新一篇:dlmalloc、nedmalloc

  • 发表评论
  • 表 情:
  • 评论内容:
  •  
Posted on 2010-10-03 13:52 袁斌 阅读(811) 评论(0)  编辑 收藏 引用

只有注册用户登录后才能发表评论。
网站导航: 博客园   IT新闻   BlogJava   博问   Chat2DB   管理