实现高并发内存池（C++）

什么是内存池

池化技术

所谓"池化技术"，就是程序先向系统申请过量的资源，然后自己管理以备不时之需。之所以要申请过量的资源，是因为每次申请该资源都有较大的开销，不如提前申请好，这样使用时就会变得非常快捷，大大提高程序运行效率。在计算机中，有很多使用"池"这种技术的地方，除了内存池，还有连接池、线程池、对象池等。

以服务器上的线程池为例，它的主要思想是：先启动若干数量的线程，让它们处于睡眠状态，当接收到客户端的请求时，唤醒池中某个睡眠的线程，让它来处理客户端的请求，当处理完这个请求，线程又进入睡眠状态。

内存池

内存池是指程序预先从操作系统申请一块足够大内存，此后，当程序中需要申请内存的时候，不是直接向操作系统申请，而是直接从内存池中获取；同理，当程序释放内存的时候，并不真正将内存返回给操作系统，而是返回内存池。当程序退出(或者特定时间)时，内存池才将之前申请的内存真正释放。

内存池主要解决的问题

内存池主要解决的当然是效率的问题，其次,作为系统的内存分配器的角度，还需要解决一下内存碎片的问题。

内存碎片

内存碎片分为外碎片和内碎片

外部碎片是一些空闲的连续内存区域太小，这些内存空间不连续，以至于合计的内存足够，但是不能满足一些的内存分配申请需求。
内部碎片是由于一些对齐的需求，导致分配出去的空间中一些内存无法被利用。

malloc

C/C++中我们要动态申请内存都是通过malloc去申请内存，实际我们不是直接去堆获取内存的。

而malloc就是一个内存池。malloc() 相当于向操作系统申请了一块较大的内存空间。当内存用完或程序有大量的内存需求时，再根据实际需求向操作系统"申请。

定长内存池

申请内存使用的是malloc，什么场景下都可以用，但是意味着什么场景下都不会有很高的性能，下面我们就先来设计一个定长内存池

ObjectPool.h

cpp 复制代码

#pragma once
#include <iostream>
#include <vector>
#include <time.h>

using std::cout;
using std::endl;
//定长内存池

//template <size_t N>
//class ObjectPool
//{};
#ifdef _WIN32
	#include <windows.h>
#else
	
#endif

inline static void* SystemAlloc(size_t kpage)//直接去堆上按页申请内存
{
#ifdef _WIN32
	void* ptr = VirtualAlloc(0, kpage<<13, MEM_COMMIT | MEM_RESERVE,
		PAGE_READWRITE);
#else
	// linux下brk mmap等
#endif
	if (ptr == nullptr)
		throw std::bad_alloc();
	return ptr;
}
template <class T>
class ObjectPool
{
public:
	T* New()
	{
		T* obj = nullptr;
		if (_freeList)
		{
			//优先把还回来的内存块再次重复利用
			void* next = (*(void**)_freeList);
			obj = (T*)_freeList;
			_freeList = next;
		}
		else
		{
			//剩余内存不够一个对象大小时，重新开大块空间
			if (remainBytes < sizeof(T))
			{
				remainBytes = 128 * 1024 ;
				//_memory = (char*)malloc(remainBytes);
				_memory = (char*)SystemAlloc(remainBytes >> 13);
				if (_memory == nullptr)
				{
					throw std::bad_alloc();
				}
			}
			obj = (T*)_memory;
			size_t objSize = sizeof(T) < sizeof(void*) ? sizeof(void*) : sizeof(T);
			_memory += objSize;
			remainBytes -= objSize;			
		}
		//定位new,显示调用T的构造函数初始化，对已有的空间初始化
		new(obj)T;

		return obj;

		
	}
	void Delete(T* obj)
	{
		
		//还回来
		
		//显示调用析构函数清理对象
		obj->~T();
		if (_freeList == nullptr)
		{
			_freeList = obj;
			//*(int*)obj = nullptr;//前四个字节用来保存下一个内存的地址 把obj强转成int* 再解引用->int 获得此地址 64位下跑不了
			*(void**)obj = nullptr;//64位下解引用是void *,*(int**)也可以
		}
		else
		{
			//头插
			*(void**)obj = _freeList;
			_freeList = obj;
		}
	}
private:
	char* _memory = nullptr;//指向大块内存，char是一个字节，好切分内存
	size_t remainBytes = 0;//大块内存中剩余数
	void* _freeList = nullptr;//管理换回来的内存(链表)的头指针
	
};

高并发内存池整体框架

现代很多的开发环境都是多核多线程，在申请内存的场景下，必然存在激烈的锁竞争问题。

内存池需要考虑以下几方面的问题。

性能问题。
多线程环境下，锁竞争问题。
内存碎片问题。

concurrent memory pool：

**thread cache：**线程缓存是每个线程独有的，用于小于256KB的内存的分配，线程从这里申请内存不需要加锁，每个线程独享一个cache，这也就是这个并发线程池高效的地方。
**central cache：**中心缓存是所有线程所共享，thread cache是按需从central cache中获取的对象。central cache合适的时机回收thread cache中的对象，避免一个线程占用了太多的内存，而其他线程的内存吃紧，达到内存分配在多个线程中更均衡的按需调度的目的。central cache是存在竞争的，所以从这里取内存对象是需要加锁，首先这里用的是桶锁，其次只有thread cache的没有内存对象时才会找central cache，所以这里竞争不会很激烈。
**page cache：**页缓存是在central cache缓存上面的一层缓存，存储的内存是以页为单位存储及分配的，central cache没有内存对象时，从page cache分配出一定数量的page，并切割成定长大小的小块内存，分配给central cache。当一个span的几个跨度页的对象都回收以后，page cache会回收central cache满足条件的span对象，并且合并相邻的页，组成更大的页，缓解内存碎片的问题。

thread cache

thread cache是哈希桶结构，每个桶是一个按桶位置映射大小的内存块对象的自由链表。每个线程都会有一个thread cache对象，这样每个线程在这里获取对象和释放对象时是无锁的。

自由链表的哈希桶跟对象大小的映射关系

class SizeClass//计算对象大小的对齐映射规则
{
public:
	// 整体控制在最多10%左右的内碎片浪费
	// [1,128] 8byte对齐       freelist[0,16)
	// [128+1,1024] 16byte对齐   freelist[16,72)
	// [1024+1,8*1024] 128byte对齐   freelist[72,128)
	// [8*1024+1,64*1024] 1024byte对齐     freelist[128,184)
	// [64*1024+1,256*1024] 8*1024byte对齐   freelist[184,208)
	static inline size_t _RoundUp(size_t bytes, size_t alignNum)//计算对齐数
	{
		//size_t alignSize ;//对齐
		//if (size % 8 != 0)
		//{
		//	alignSize = (size / alignNum + 1) * alignNum;
		//}
		//else
		//{
		//	alignSize = size;
		//}
		//return alignSize;
		return ((bytes + alignNum - 1) & ~(alignNum - 1));
	}
	static inline size_t RoundUp(size_t size)
	{
		if (size <= 128)
		{
			return _RoundUp(size, 8);
		}
		else if (size <= 1024)
		{
			return _RoundUp(size, 16);
		}
		else if (size <= 8 * 1024)
		{
			return _RoundUp(size, 128);
		}
		else if (size <= 64 * 1024)
		{
			return _RoundUp(size, 1024);
		}
		else if (size <= 256 * 1024)
		{
			return _RoundUp(size, 8 * 1024);
		}
		else
		{
			return _RoundUp(size, 1 << PAGE_SHIFT);
		}
	}

映射哪一个自由链表桶

static inline size_t _Index(size_t bytes, size_t alignNum)
	{
		/*if (bytes % alignNum == 0)
		{
		return bytes / alignNum - 1;
		}
		else
		{
		return bytes / alignNum;
		}*/
		return ((bytes + (1 << alignNum) - 1) >> alignNum) - 1;
	}
	// 计算映射的哪一个自由链表桶
	static inline size_t Index(size_t bytes)
	{
		assert(bytes <= MAX_BYTES);
		// 每个区间有多少个链
		static int group_array[4] = { 16, 56, 56, 56 };
		if (bytes <= 128) {
			return _Index(bytes, 3);//8  2^3
		}
		else if (bytes <= 1024) {
			return _Index(bytes - 128, 4) + group_array[0];//把前面128减掉，再加上前一个桶的数量
		}
		else if (bytes <= 8 * 1024) {
			return _Index(bytes - 1024, 7) + group_array[1] + group_array[0];
		}
		else if (bytes <= 64 * 1024) {
			return _Index(bytes - 8 * 1024, 10) + group_array[2] + group_array[1] + group_array[0];
		}
		else if (bytes <= 256 * 1024) {
			return _Index(bytes - 64 * 1024, 13) + group_array[3] + group_array[2] + group_array[1] + group_array[0];
		}
		else {
			assert(false);
		}
		return -1;
	}

申请内存：

当内存申请size<=256KB，先获取到线程本地存储的thread cache对象，计算size映射的哈希桶自由链表下标i。
如果自由链表_freeLists[i]中有对象，则直接Pop一个内存对象返回。
如果_freeLists[i]中没有对象时，则批量从central cache中获取一定数量的对象，插入到自由链表并返回一个对象。

释放内存

当释放内存小于256k时将内存释放回thread cache，计算size映射自由链表桶位置i，将对象Push到_freeLists[i]。
当链表的长度过长，则回收一部分内存对象到central cache。

thread cache 设计

#pragma once
#include "Common.h"

class ThreadCache
{
public:
	//申请和释放对象
	void* Allocate(size_t size);
	void Deallocate(void* ptr, size_t size);
	//从中心缓存获取对象
	void* FetchFromCentralCache(size_t index, size_t size);

	void ListTooLong(FreeList& list, size_t size);//释放对象时，链表过长  ，回收内存到centrral cache
private:
	FreeList _freeLists[NFREELISTS];//哈希表，每个位置挂的都是_freeList
};

// TLS:在线程内全局可访问，但不能被其他线程访问到->保持数据的独立性，不需要锁控制，减少成本
static _declspec(thread) ThreadCache * pTLSThreadCache = nullptr;

central cache

central cache也是一个哈希桶结构(t桶锁)，他的哈希桶的映射关系跟thread cache是一样的。不同的是他的每个哈希桶位置挂是SpanList链表结构，不过每个映射桶下面的span中的大内存块被按映射关系切成了一个个小内存块对象挂在span的自由链表中。

申请内存：

当thread cache中没有内存时，就会批量向central cache申请一些内存对象，这里的批量获取对象的数量使用了类似网络tcp协议拥塞控制的慢开始算法；central cache也有一个哈希映射的spanlist，spanlist中挂着span，从span中取出对象给thread cache，这个过程是需要加锁的，这里使用的是一个桶锁，尽可能提高效率。
central cache映射的spanlist中所有span的都没有内存以后，则需要向page cache申请一个新的span对象，拿到span以后将span管理的内存按大小切好作为自由链表链接到一起。然后从span中取对象给thread cache
central cache的中挂的span中use_count记录分配了多少个对象出去，分配一个对象给threadcache，就++use_count

释放内存

当thread_cache过长或者线程销毁，则会将内存释放回central cache中的，释放回来时--use_count。当use_count减到0时则表示所有对象都回到了span，则将span释放回page cache，page cache中会对前后相邻的空闲页进行合并。

以页为单位的大内存管理span的定义及spanlist定义

struct Span//管理多个连续大块内存跨度结构
{
	PAGE_ID _pageId = 0;//大块内存起始页号
	size_t n = 0;//页的数量
	Span* _next = nullptr;//双向链表
	Span* _prev = nullptr;//双向链表

	size_t _objSize = 0;//切好的小对象的大小
	size_t _useCount = 0;//切好的小块内存，被分给thread cache计数
	void* _freeList = nullptr;//切好的小块内存自由链表
	bool _isUse = false;//是否被使用

};
class SpanList//带头双向链表
{
public:
	SpanList()
	{
		_head = new Span;
		_head->_next = _head;
		_head->_prev = _head;
	}

	Span* Begin()
	{
		return _head->_next;
	}
	Span* End()
	{
		return _head;
	}
	bool Empty()
	{
		//cout << "heool spanlist empty" << endl;
		return _head->_next == _head;
	}
	void PushFront(Span* span)
	{
		//cout << "hello common pushfront" << endl;
		Insert(Begin(), span);
	}
	Span* PopFront()
	{
		//cout << "hello commom popfront" << endl;
		Span* front = _head->_next;
		Erase(front);
		return front;
	}
	void Insert(Span* pos, Span* newSpan)
	{
		//cout << "hello commom insert" << endl;
		assert(pos);
		assert(newSpan);
		Span* prev = pos->_prev;
		prev->_next = newSpan;
		newSpan->_prev = prev;
		newSpan->_next = pos;
		pos->_prev = newSpan;
	}
	void Erase(Span* pos)
	{
		assert(pos);
		assert(pos != _head);

		Span* prev = pos->_prev;
		Span* next = pos->_next;
		prev->_next = next;
		next->_prev = prev;

	}
private:
	Span* _head;
public:

	std::mutex _mtx;//桶锁

};

central cache整体设计

#pragma once
#include "Common.h"

//单例模式
class CentralCache
{
public:
	static CentralCache* GetInstance()
	{
		return &_sInst;
	}

	// 获取一个非空的span
	Span* GetOneSpan(SpanList& list, size_t byte_size);

	// 从中心缓存获取一定数量的对象给thread cache
	size_t FetchRangeObj(void*& start, void*& end, size_t batchNum, size_t size);

	// 将一定数量的对象释放到span跨度
	void ReleaseListToSpans(void* start, size_t byte_size);

private:
	SpanList _spanLists[NFREELISTS];//在ThreadCache是几号桶，在CentralCache就是几号桶
private:
	CentralCache() //把构造函数放在私有：别人不能创建对象
	{}
	CentralCache(const CentralCache&) = delete;
	static CentralCache _sInst;

};

page cache

申请内存：

当central cache向page cache申请内存时，page cache先检查对应位置有没有span，如果没有则向更大页寻找一个span，如果找到则分裂成两个。比如：申请的是4页page，4页page后面没有挂span，则向后面寻找更大的span，假设在10页page位置找到一个span，则将10页page span分裂为一个4页pagespan和一个6页page span
如果找到_spanList[128]都没有合适的span，则向系统使用mmap、brk或者是VirtualAlloc等方式申请128页page span挂在自由链表中，再重复1中的过程。
需要注意的是central cache和page cache 的核心结构都是spanlist的哈希桶，但是他们是有本质区别的，central cache中哈希桶，是按跟thread cache一样的大小对齐关系映射的，他的spanlist中挂的span中的内存都被按映射关系切好链接成小块内存的自由链表。而page cache 中的spanlist则是按下标桶号映射的，也就是说第i号桶中挂的span都是i页内存。

释放内存：

如果central cache释放回一个span，则依次寻找span的前后page id的没有在使用的空闲span，看是否可以合并，如果合并继续向前寻找。这样就可以将切小的内存合并收缩成大的span，减少内存碎片
如果Central Cache中的span usecount=0说明切分给 thread cache的小块内存都回来了则Central Cache 把这个span还给page cache,page cache通过页号查看前后相邻页是否空闲，是就合并出更大的页

整体设计

#pragma once
#include "Common.h"
#include "ObjectPool.h"

class PageCache
{
public:
	static PageCache* GetInstance()
	{
		//cout << "hello Page cache getinstance" << endl;
		return &_sInst;
	}
	Span* MapObjectToSpan(void* obj);//获取对象到span的映射

	Span* NewSpan(size_t k);//获取一个k页的span

	void ReleaseSpanToPageCache(Span* span);//释放空闲span,合并相邻的span
	std::mutex _pageMtx;
private:
	SpanList _spanList[NPAGES];
	ObjectPool<Span> spanPool;
	std::unordered_map<PAGE_ID,Span*> _idSpanMap;//页号跟span的映射
	PageCache() {}
	PageCache(const PageCache&) = delete;
	static PageCache _sInst;
};

代码总体实现

ObjectPool.h

#pragma once
#pragma once
#include <iostream>
#include <vector>
#include <time.h>
#include "Common.h"
using std::cout;
using std::endl;
//定长内存池

//template <size_t N>
//class ObjectPool
//{};

/*
#ifdef _WIN32
#include <windows.h>
#else

#endif
inline static void* SystemAlloc(size_t kpage)//直接去堆上按页申请内存
{
#ifdef _WIN32
	void* ptr = VirtualAlloc(0, kpage << 13, MEM_COMMIT | MEM_RESERVE,
		PAGE_READWRITE);
#else
	// linux下brk mmap等
#endif
	if (ptr == nullptr)
		throw std::bad_alloc();
	return ptr;
}
*/
template <class T>
class ObjectPool
{
public:
	T* New()
	{
		T* obj = nullptr;
		if (_freeList)
		{
			//优先把还回来的内存块再次重复利用
			void* next = (*(void**)_freeList);
			obj = (T*)_freeList;
			_freeList = next;
			
		}
		else
		{
			//剩余内存不够一个对象大小时，重新开大块空间
			if (remainBytes < sizeof(T))
			{
				remainBytes = 128 * 1024;
				//_memory = (char*)malloc(remainBytes);
				_memory = (char*)SystemAlloc(remainBytes >> 13);
				if (_memory == nullptr)
				{
					throw std::bad_alloc();
				}
			}
			obj = (T*)_memory;
			size_t objSize = sizeof(T) < sizeof(void*) ? sizeof(void*) : sizeof(T);
			_memory += objSize;
			remainBytes -= objSize;
		}
		//定位new,显示调用T的构造函数初始化，对已有的空间初始化
		new(obj)T;

		return obj;


	}
	void Delete(T* obj)
	{

		//还回来

		//显示调用析构函数清理对象
		obj->~T();
		if (_freeList == nullptr)
		{
			_freeList = obj;
			//*(int*)obj = nullptr;//前四个字节用来保存下一个内存的地址 把obj强转成int* 再解引用->int 获得此地址 64位下跑不了
			*(void**)obj = nullptr;//64位下解引用是void *,*(int**)也可以
		}
		else
		{
			//头插
			*(void**)obj = _freeList;
			_freeList = obj;
		}
	}
private:
	char* _memory = nullptr;//指向大块内存，char是一个字节，好切分内存
	size_t remainBytes = 0;//大块内存中剩余数
	void* _freeList = nullptr;//管理换回来的内存(链表)的头指针

};

Common.h

#pragma once
//公共文件

#include <iostream>
#include <vector>
#include <time.h>
#include <assert.h>
#include <thread>
#include <mutex>
#include <algorithm>
#include <windows.h>
#include <unordered_map>
#include <map>
using std::cout;
using std::endl;

static const size_t MAX_BYTES = 256 * 1024;//256 KB
static const size_t NFREELISTS = 208;//桶的总数量
static const size_t NPAGES = 129;//页的数量
static const size_t PAGE_SHIFT = 13;//

#ifdef _WIN64
typedef unsigned long long PAGE_ID;
#elif _WIN32
typedef size_t PAGE_ID;
#else
 //linux
#endif

inline static void* SystemAlloc(size_t kpage)//直接去堆上按页申请内存
{
#ifdef _WIN32
	void* ptr = VirtualAlloc(0, kpage << 13, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
#else
	// linux下brk mmap等
#endif
	if (ptr == nullptr)
		throw std::bad_alloc();
	return ptr;
}
inline static void SystemFree(void* ptr)
{
#ifdef _WIN32
	VirtualFree(ptr, 0, MEM_RELEASE);
#else
	// sbrk unmmap等
#endif
}

static void*& NextObj(void* obj)//取对象的头4/8字节
{
	return *(void**)obj;
}
class FreeList //管理切分好的小对象的自由链表
{
public:
	void Push(void* obj)//插入
	{
		//头插
		//*(void**)obj = _freeList;
		NextObj(obj) = _freeList;
		_freeList = obj;
		_size++;
	}
	void PushRange(void* start, void* end, size_t n)
	{
		//cout << "hello common pushrange" << endl;
		NextObj(end) = _freeList;
		_freeList = start;

		/*
		//测试验证+条件断点
		int i = 0;
		void* cur = start;
		while (cur)
		{
			i++;
			cur = NextObj(cur);
		}
		if (n != i)
		{
			//int x = 0;
			cout << "不匹配" << endl;
		}
		*/
		_size += n;

	}
	void PopRange(void*& start, void*& end, size_t n)
	{
		
		assert(n >= _size);
		start = _freeList;
		end = start;
		for (size_t i = 0; i < n - 1; i++)
		{
			end = NextObj(end);
		}
		_freeList = NextObj(end);
		NextObj(end) = nullptr;
		_size -= n;
	}
	void* Pop()//弹出对象
	{
		//头删
		assert(_freeList);
		void* obj = _freeList;
		_freeList = NextObj(obj);
		_size--;
		return obj;
	}
	bool Empty()
	{
		return _freeList == nullptr;
	}
	size_t& MaxSize()
	{
		return _maxSize;
	}
	size_t Size()
	{
		return _size;
	}
private:
	void* _freeList = nullptr;
	size_t _maxSize = 1;//
	size_t _size = 0;//个数
};

class SizeClass//计算对象大小的对齐映射规则
{
public:
	// 整体控制在最多10%左右的内碎片浪费
	// [1,128] 8byte对齐       freelist[0,16)
	// [128+1,1024] 16byte对齐   freelist[16,72)
	// [1024+1,8*1024] 128byte对齐   freelist[72,128)
	// [8*1024+1,64*1024] 1024byte对齐     freelist[128,184)
	// [64*1024+1,256*1024] 8*1024byte对齐   freelist[184,208)
	static inline size_t _RoundUp(size_t bytes, size_t alignNum)//计算对齐数
	{
		//size_t alignSize ;//对齐
		//if (size % 8 != 0)
		//{
		//	alignSize = (size / alignNum + 1) * alignNum;
		//}
		//else
		//{
		//	alignSize = size;
		//}
		//return alignSize;
		return ((bytes + alignNum - 1) & ~(alignNum - 1));
	}
	static inline size_t RoundUp(size_t size)//计算对齐数
	{
		if (size <= 128)
		{
			return _RoundUp(size, 8);
		}
		else if (size <= 1024)
		{
			return _RoundUp(size, 16);
		}
		else if (size <= 8 * 1024)
		{
			return _RoundUp(size, 128);
		}
		else if (size <= 64 * 1024)
		{
			return _RoundUp(size, 1024);
		}
		else if (size <= 256 * 1024)
		{
			return _RoundUp(size, 8 * 1024);
		}
		else
		{
			return _RoundUp(size, 1 << PAGE_SHIFT);
			//assert(false);
			//return -1;
		}
	}

	static inline size_t _Index(size_t bytes, size_t alignNum)
	{
		/*if (bytes % alignNum == 0)
		{
		return bytes / alignNum - 1;
		}
		else
		{
		return bytes / alignNum;
		}*/
		return ((bytes + (1 << alignNum) - 1) >> alignNum) - 1;
	}
	// 计算映射的哪一个自由链表桶
	static inline size_t Index(size_t bytes)
	{
		assert(bytes <= MAX_BYTES);
		// 每个区间有多少个链
		static int group_array[4] = { 16, 56, 56, 56 };
		if (bytes <= 128) {
			return _Index(bytes, 3);//8  2^3
		}
		else if (bytes <= 1024) {
			return _Index(bytes - 128, 4) + group_array[0];//把前面128减掉，再加上前一个桶的数量
		}
		else if (bytes <= 8 * 1024) {
			return _Index(bytes - 1024, 7) + group_array[1] + group_array[0];
		}
		else if (bytes <= 64 * 1024) {
			return _Index(bytes - 8 * 1024, 10) + group_array[2] + group_array[1] + group_array[0];
		}
		else if (bytes <= 256 * 1024) {
			return _Index(bytes - 64 * 1024, 13) + group_array[3] + group_array[2] + group_array[1] + group_array[0];
		}
		else {
			assert(false);
		}
		return -1;
	}
	static size_t NumMoveSize(size_t size)// 一次thread cache从中心缓存获取多少个对象
	{
		assert(size > 0);
		// [2, 512]，一次批量移动多少个对象的(慢启动)上限值
		// 小对象一次批量上限高
		// 小对象一次批量上限低
		int num = MAX_BYTES / size;
		if (num < 2)
			num = 2;
		if (num > 512)
			num = 512;
		return num;
	}


	static size_t NumMovePage(size_t size)// 计算一次向系统获取几个页
	{
		// 单个对象 8byte
		// ...
		// 单个对象 256KB
		size_t num = NumMoveSize(size);
		size_t npage = num * size;
		npage >>= PAGE_SHIFT;
		if (npage == 0)
			npage = 1;
		return npage;
	}
};

struct Span//管理多个连续大块内存跨度结构
{
	PAGE_ID _pageId = 0;//大块内存起始页号
	size_t n = 0;//页的数量
	Span* _next = nullptr;//双向链表
	Span* _prev = nullptr;//双向链表

	size_t _objSize = 0;//切好的小对象的大小
	size_t _useCount = 0;//切好的小块内存，被分给thread cache计数
	void* _freeList = nullptr;//切好的小块内存自由链表
	bool _isUse = false;//是否被使用

};
class SpanList//带头双向链表
{
public:
	SpanList()
	{
		_head = new Span;
		_head->_next = _head;
		_head->_prev = _head;
	}

	Span* Begin()
	{
		return _head->_next;
	}
	Span* End()
	{
		return _head;
	}
	bool Empty()
	{
		//cout << "heool spanlist empty" << endl;
		return _head->_next == _head;
	}
	void PushFront(Span* span)
	{
		//cout << "hello common pushfront" << endl;
		Insert(Begin(), span);
	}
	Span* PopFront()
	{
		//cout << "hello commom popfront" << endl;
		Span* front = _head->_next;
		Erase(front);
		return front;
	}
	void Insert(Span* pos, Span* newSpan)
	{
		//cout << "hello commom insert" << endl;
		assert(pos);
		assert(newSpan);
		Span* prev = pos->_prev;
		prev->_next = newSpan;
		newSpan->_prev = prev;
		newSpan->_next = pos;
		pos->_prev = newSpan;
	}
	void Erase(Span* pos)
	{
		assert(pos);
		//assert(pos != _head);

		//1、条件断点
		//2、查看栈帧
		/*
		if (pos = _head)
		{
			int x = 0;
		}
		*/
		Span* prev = pos->_prev;
		Span* next = pos->_next;
		prev->_next = next;
		next->_prev = prev;

	}
private:
	Span* _head;
public:

	std::mutex _mtx;//桶锁

};

ConcurrentAlloc.h

#pragma once
#include "Common.h"
#include "ThreadCache.h"
#include "PageCache.h"
#include "ObjectPool.h"
static void* ConcurrentAlloc(size_t  size)//线程调用申请内存
{
	//通过TLS 每个线程无锁的获取自己的专属ThreadCache对象
	if (size > MAX_BYTES)
	{
		size_t alignSize = SizeClass::RoundUp(size);//对齐
		size_t kpage = alignSize >> PAGE_SHIFT;//获取页数

		PageCache::GetInstance()->_pageMtx.lock();
		Span* span = PageCache::GetInstance()->NewSpan(kpage);
		//span->_objSize = size;
		PageCache::GetInstance()->_pageMtx.unlock();

		void* ptr = (void*)(span->_pageId << PAGE_SHIFT);
		return ptr;
	}
	else
	{
		if (pTLSThreadCache == nullptr)
		{
			
			//pTLSThreadCache = new ThreadCache;
			static ObjectPool<ThreadCache> tcPool;
			pTLSThreadCache = tcPool.New();
		}
		cout << std::this_thread::get_id() << ":" << pTLSThreadCache << endl;
		return pTLSThreadCache->Allocate(size);
	}
	
}
static void ConcurrentFree(void* ptr)
{
	//size:不给大小不知道要还给桶的哪个位置
	Span* span = PageCache::GetInstance()->MapObjectToSpan(ptr);
	size_t size = span->_objSize;//对齐以后的大小
	if (size > MAX_BYTES)
	{
	
		

		PageCache::GetInstance()->_pageMtx.lock();
		
		
		PageCache::GetInstance()->ReleaseSpanToPageCache(span);
		PageCache::GetInstance()->_pageMtx.unlock();
	}
	else
	{
		assert(pTLSThreadCache);
		pTLSThreadCache->Deallocate(ptr, size);
	}

	
}

ThreadCache.h

#pragma once
#include "Common.h"

class ThreadCache
{
public:
	//申请和释放对象
	void* Allocate(size_t size);
	void Deallocate(void* ptr, size_t size);
	//从中心缓存获取对象
	void* FetchFromCentralCache(size_t index, size_t size);

	void ListTooLong(FreeList& list, size_t size);//释放对象时，链表过长  ，回收内存到centrral cache
private:
	FreeList _freeLists[NFREELISTS];//哈希表，每个位置挂的都是_freeList
};

// TLS:在线程内全局可访问，但不能被其他线程访问到->保持数据的独立性，不需要锁控制，减少成本
static _declspec(thread) ThreadCache * pTLSThreadCache = nullptr;

ThreadCache.cpp

#include "ThreadCache.h"
#include "CentralCache.h"
#include "Common.h"

void* ThreadCache::FetchFromCentralCache(size_t index, size_t size)
{
	cout << "hello common fecthcenrercache" << endl;
	//慢开始反馈调节算法
	//最开始不会向 central cache要太多因为可能用不完，如果不要size大小需求batchNum会不断增长直到上限;
	//size越大一次向central cache要的越小，size越小一次向central cache要的越大
	size_t batchNum = min(_freeLists[index].MaxSize(), SizeClass::NumMoveSize(size));

	if (_freeLists[index].MaxSize() == batchNum)
	{
		_freeLists[index].MaxSize() += 1;
	}

	void* start = nullptr;
	void* end = nullptr;
	size_t actualNum = CentralCache::GetInstance()->FetchRangeObj(start, end, batchNum,size);

	

	assert(actualNum > 0);
	if (actualNum == 1)
	{
		assert(start == end);
		return start;
	}
	else
	{
		_freeLists[index].PushRange(NextObj(start), end, actualNum - 1);
		return start;
	}

	return nullptr;
}
void* ThreadCache::Allocate(size_t size)//申请对象
{
	//
	assert(size <= MAX_BYTES);
	size_t alignSize = SizeClass::RoundUp(size);//获取对其数
	size_t index = SizeClass::Index(size);//在哪一个桶-》获取桶的位置
	if (!_freeLists[index].Empty())
	{
		return _freeLists[index].Pop();
	}
	else
	{
		return FetchFromCentralCache(index,alignSize);//从中心缓存获取对象
	}
}
void ThreadCache::Deallocate(void* ptr, size_t size)//释放对象
{
	assert(size <= MAX_BYTES);
	assert(ptr);

	//找出自由链表映射的桶，对象插入
	size_t index = SizeClass::Index(size);//属于哪个桶
	_freeLists[index].Push(ptr);

	//当链表长度大于等于一次批量申请的内存时就开始还一段内存给central cache
	if (_freeLists[index].Size() >= _freeLists[index].MaxSize())
	{	ListTooLong(_freeLists[index], size);
	}
}
void ThreadCache::ListTooLong(FreeList& list, size_t size)
{
	void* start = nullptr;
	void* end = nullptr;
	list.PopRange(start,end,list.MaxSize());//取出内存

	//把内存还给下一层：central cache
	CentralCache::GetInstance()->ReleaseListToSpans(start,size);
}

CentralCache.h

#include "CentralCache.h"
#include "PageCache.h"
CentralCache CentralCache::_sInst;
Span* CentralCache::GetOneSpan(SpanList& list, size_t size)
{
	//cout << "hello central getonspan" << endl;
	
	// 从SpanLists或者Page cache 获取一个非空的span
	//查看当前spanlist中是否还有非空的/还未分配对象的
	Span* it = list.Begin();
	
	while (it != list.End())
	{
		if (it->_freeList != nullptr)
		{
			//挂着对象
			return it;
		}
		else
		{
			it = it->_next;
		}

	}
	
	//先把central cache的桶锁解掉，这样如果其他线程释放内存对象回来不会阻塞
	list._mtx.unlock();

	//没有空闲span,找 page cache要
	PageCache::GetInstance()->_pageMtx.lock();
	Span* span = PageCache::GetInstance()->NewSpan(SizeClass::NumMovePage(size));
	span->_isUse = true;
	span->_objSize = size;
	PageCache::GetInstance()->_pageMtx.unlock();

	//对获取的span进行切分吧、，不需要加锁，因为其他线程访问不到这个span

	//通过页号计算起始地址: 页号<<PAGE_SHIFT
	char* start = (char*)(span->_pageId << PAGE_SHIFT);
	size_t bytes = span->n << PAGE_SHIFT;//计算span的大块起始地址和大块内存的大小(字节数)
	char* end = start + bytes;

	//把大块内存切成自由链表连接起来
	//先切一块下来去做头，方便尾插
	span->_freeList = start;
	start += size;
	int i = 1;
	void* tail = span->_freeList;
	while (start < end)
	{
		i++;
		NextObj(tail) = start;
		tail = NextObj(tail);//tail = start
		start += size;
	}
	NextObj(tail) = nullptr;//尾插最后一位需要置空

	list._mtx.lock();//切好span以后需要把span挂到桶里去再加锁
	list.PushFront(span);

	return span;
}

size_t CentralCache::FetchRangeObj(void*& start, void*& end, size_t batchNum, size_t size)// 从中心缓存获取一定数量的对象给thread cache
{
	//cout << "hello hello central getonspan" << endl;
	size_t index = SizeClass::Index(size);//先查看是哪个桶的
	_spanLists[index]._mtx.lock();

	Span* span = GetOneSpan(_spanLists[index], size);//先去找一个非空的span
	assert(span);
	assert(span->_freeList);

	//从span中获取batchNum个对象
	//如果不够batchNum,有多少拿多少

	start = span->_freeList;
	end = start;
	size_t i = 0;
	size_t actualNum = 1;
	while (i < batchNum - 1 && NextObj(end) != nullptr)//end 往后走batchNum -1个
	{
		end = NextObj(end);
		++i;
		++actualNum;
	}

	span->_freeList = NextObj(end);
	NextObj(end) = nullptr;
	span->_useCount += actualNum;//被使用的个数 

	_spanLists[index]._mtx.unlock();
	return actualNum;
	

}
void CentralCache::ReleaseListToSpans(void* start, size_t size)
{
	size_t index = SizeClass::Index(size);//属于哪一个桶
	_spanLists[index]._mtx.lock();
	while (start)
	{
		void* next = NextObj(start);
		Span* span = PageCache::GetInstance()->MapObjectToSpan(start);//找出对应的span
		NextObj(start) = span->_freeList;
		span->_freeList = start;
		span->_useCount--;
		if (span->_useCount == 0)//说明span切分出去的的所有小内存都回来了，
		{//该span可以归还给page cache,page cache再可以去做前后页的合并
			_spanLists[index].Erase(span);
			span->_freeList = nullptr;
			span->_next = nullptr;
			span->_prev = nullptr;

			//span还给下一层
			_spanLists[index]._mtx.unlock();

			//释放span给Page cache时，使用page cache锁
			//这时把桶锁解掉
			PageCache::GetInstance()->_pageMtx.lock();
			PageCache::GetInstance()->ReleaseSpanToPageCache(span);
			PageCache::GetInstance()->_pageMtx.unlock();

			_spanLists[index]._mtx.lock();
		}
		start = next;
	}


	_spanLists[index]._mtx.unlock();


}

Pagecache.h

#pragma once
#include "Common.h"
#include "ObjectPool.h"

class PageCache
{
public:
	static PageCache* GetInstance()
	{
		//cout << "hello Page cache getinstance" << endl;
		return &_sInst;
	}
	Span* MapObjectToSpan(void* obj);//获取对象到span的映射

	Span* NewSpan(size_t k);//获取一个k页的span

	void ReleaseSpanToPageCache(Span* span);//释放空闲span,合并相邻的span
	std::mutex _pageMtx;
private:
	SpanList _spanList[NPAGES];
	ObjectPool<Span> spanPool;
	std::map<PAGE_ID, Span*> _idSpanMap;//页号跟span的映射
	//std::unordered_map<PAGE_ID,Span*> _idSpanMap;//页号跟span的映射

	PageCache() {}
	PageCache(const PageCache&) = delete;
	static PageCache _sInst;
};

PageCache.cpp

#include "PageCache.h"
PageCache PageCache::_sInst;

Span* PageCache::NewSpan(size_t k)//获取k页的span
{
	//eg:只有一个128页的，需要两页的-》128分为2span和126span，2返回给central cache,126挂道对应的桶上
	//如果central cache中的span usecount=0,说明切分给thread cache小块内存都还回来了，
	//则central cache把span还给page cache,page cache通过页号查看相邻页是否空闲，是就合并出更大的page,解决内存碎片问题
	assert(k > 0 && k < NPAGES);
	
	if (k > NPAGES - 1)
	{
		void* ptr = SystemAlloc(k);//大于最大页数直接找堆要
		//Span* span = new Span;
		Span* span = spanPool.New();
		span->_pageId = (PAGE_ID)ptr >> PAGE_SHIFT;
		span->n = k;
		_idSpanMap[span->_pageId] = span;
		return span;
	}
	
	if (!_spanList[k].Empty())//第k个桶里面有没有span
	{
		
		Span* KSpan = _spanList[k].PopFront();
		
		//建立id和span的映射关系方便centralcache回收小块内存时查找对应的span
		for (PAGE_ID i = 0; i < KSpan->n; i++)
		{
			_idSpanMap[KSpan->_pageId + i] = KSpan;

		}
		
		return KSpan;
	}

	//第k个桶里是空的,检测后面的桶里有没有span,如果有进行切分
	//切分成一个k页的span和一个 n-k 页的span
	//k页的span返回给central cache,n-k 页的span挂到第 n-k 号桶中去
	for (size_t i = k ; i < NPAGES; i++)
	{		
		if (!_spanList[i].Empty())
		{
			Span* nSpan = _spanList[i].PopFront();
			//Span* KSpan = new Span;
			Span* KSpan = spanPool.New();
			

			//在nSpan的头部切下K页
			//k页span返回，nSpan再挂到对应映射
			KSpan->_pageId = nSpan->_pageId;
			KSpan->n = k;//kSpan页数变为k
			nSpan->_pageId += k;//nSpan 页号变为 += k
			nSpan->n -= k;
			
			_spanList[nSpan->n].PushFront(nSpan);//把剩余的页挂到对应的位置

			
			//存储nSpan的首尾页号跟span映射,方便page cache回收内存时进行合并查找
			_idSpanMap[nSpan->_pageId] = nSpan;
			_idSpanMap[nSpan->_pageId + nSpan->n - 1] = nSpan;
			

			//建立id和span的映射，方便central cache回收查找对应的span
			for (PAGE_ID i = 0; i < KSpan->n; i++)
			{
				_idSpanMap[KSpan->_pageId + i] = KSpan;
			}
			
			
			return KSpan;
		}
	}

	//没有大页span
	//找堆要128页的span
	Span* bigSpan = spanPool.New();
	//Span* bigSpan = new Span;
	void* ptr = SystemAlloc(NPAGES - 1);
	bigSpan->_pageId = (PAGE_ID)ptr >> PAGE_SHIFT;
	bigSpan->n = NPAGES - 1;
	_spanList[bigSpan->n].PushFront(bigSpan);
	return NewSpan(k);

}

Span* PageCache::MapObjectToSpan(void* obj)
{
	
	PAGE_ID id = ((PAGE_ID)obj >> PAGE_SHIFT);//找出页号
	std::unique_lock<std::mutex> lock(_pageMtx);
	auto ret = _idSpanMap.find(id);
	if (ret != _idSpanMap.end())
	{
		return ret->second;//返回span的指针
	}
	else
	{
		assert(false);
		return nullptr;
	}
}
void PageCache::ReleaseSpanToPageCache(Span* span)
{
	//对span前后的页进行合并，解决内存碎片问题

	if (span->n > NPAGES - 1)
	{
		//大于128页的直接还给堆
		void* ptr = (void*)(span->_pageId << PAGE_SHIFT);
		SystemFree(ptr);
		//delete span;
		spanPool.Delete(span);
		return;
	}
	//向前合并
	while (1)
	{
		PAGE_ID prevId = span->_pageId - 1;
		auto ret = _idSpanMap.find(prevId);

		if (ret == _idSpanMap.end())
		{//前面的页号没有，不合并
			break;
		}
		Span* prevspan = ret->second;
		if (prevspan->_isUse == true)
		{//前面相邻页的span在使用
			break;
		}
		if (prevspan->n + span->n > NPAGES - 1)
		{
			//合并数超过128，没办法管理
			break;
		}
		//合并
		span->_pageId = prevspan->_pageId;
		span->n += prevspan->n;
		_spanList[prevspan->n].Erase(prevspan);
		//delete prevspan;
		
		spanPool.Delete(prevspan);
	}

	//向后合并
	while (1)
	{
		PAGE_ID nextId = span->_pageId + span->n;
		auto ret = _idSpanMap.find(nextId);
		if (ret == _idSpanMap.end())
		{
			break;
		}
		Span* nextspan = ret->second;
		if (nextspan->_isUse == true)
		{
			break;
		}
		if (nextspan->n + span->n > NPAGES - 1)
		{
			break;
		}
		span->n += nextspan->n;
		_spanList[nextspan->n].Erase(nextspan);
		//delete(nextspan);
		spanPool.Delete(nextspan);

	}
	//前后都合并过了

	_spanList[span->n].PushFront(span);
	span->_isUse = false;

	//方便其他把此span合并
	_idSpanMap[span->_pageId] = span;
	_idSpanMap[span->_pageId + span->n - 1] = span;

}

复杂问题的调试技巧

条件断点:一般情况下可以直接运行程序，通过报错来查找问题。如果是断言错误，那么可以直接定位到报错位置，然后将此处的断言改为与if判断，在if语句里面打上一个断点(空语句是无法打断点可以随便在if里面加上一句代码)，条件断点也客设置为普通断点，设置相应的条件，程序满足该条件则会停下。
查看函数栈帧：当前函数栈帧的调用情况(黄色箭头指向的是当前所在的函数栈帧)双击函数栈帧中的其他函数可以跳转对应的栈帧(浅灰色箭头指向的就是当前跳转到的函数栈帧)
死循环时中断程序：调试→全部中断，程序会在当前运行的地方停下

vs2013性能分析

调试-》性能与诊断-》开始-》检测

实现基数树进行优化

单层基数树

实际采用的就是直接定址法，每一个页号对应span的地址就存储数组中在以该页号为下标的位置

二层基数树

这里还是以32位平台下，一页的大小为8K为例来说明，此时存储页号最多需要19个比特位。而二层基数树实际上就是把这19个比特位分为两次进行映射。

三层基数树

上面一层基数树和二层基数树都适用于32位平台，而对于64位的平台就需要用三层基数树了。三层基数树与二层基数树类似，三层基数树实际上就是把存储页号的若干比特位分为三次进行映射。

代码实现

PageMap.h

#pragma once
#include "Common.h"
#include "ObjectPool.h"
template <int BITS>
class TCMalloc_PageMap1 {
private:
	static const int LENGTH = 1 << BITS;
	void** array_;
public:
	typedef uintptr_t Number;
	explicit TCMalloc_PageMap1(void* (*allocator)(size_t)) {
		//array_ = reinterpret_cast<void**>((*allocator)(sizeof(void*) << BITS));
		size_t size = sizeof(void*) << BITS;
		size_t  alignSize = SizeClass::_RoundUp(size,1 << PAGE_SHIFT);

		array_ = SystemAlloc(alignSize >> PAGE_SHIFT);
		memset(array_, 0, sizeof(void*) << BITS);
	}
	// Return the current value for KEY. Returns NULL if not yet set,
	// or if k is out of range.
	void* get(Number k) const {
		if ((k >> BITS) > 0) {
			return NULL;
		}
		return array_[k];
	}
	// REQUIRES "k" is in range "[0,2^BITS-1]".
	// REQUIRES "k" has been ensured before.
	//
	// Sets the value 'v' for key 'k'.
	void set(Number k, void* v) {
		array_[k] = v;
	}
};
// Two-level radix tree
template <int BITS>
class TCMalloc_PageMap2 {
private:
	// Put 32 entries in the root and (2^BITS)/32 entries in each leaf.
	static const int ROOT_BITS = 5;
	static const int ROOT_LENGTH = 1 << ROOT_BITS;
	static const int LEAF_BITS = BITS - ROOT_BITS;
	static const int LEAF_LENGTH = 1 << LEAF_BITS;
	
		// Leaf node
		struct Leaf {
		void* values[LEAF_LENGTH];
	};
	Leaf* root_[ROOT_LENGTH];             // Pointers to 32 child nodes
	void* (*allocator_)(size_t);          // Memory allocator
public:
	typedef uintptr_t Number;
	//explicit TCMalloc_PageMap2(void* (*allocator)(size_t)) 
	explicit TCMalloc_PageMap2() {
		//allocator_ = allocator;
		memset(root_, 0, sizeof(root_));
	}
	void* get(Number k) const {
		const Number i1 = k >> LEAF_BITS;
		const Number i2 = k & (LEAF_LENGTH - 1);
		if ((k >> BITS) > 0 || root_[i1] == NULL) {
			return NULL;
		}
		return root_[i1]->values[i2];
	}
	void set(Number k, void* v) {
		const Number i1 = k >> LEAF_BITS;
		const Number i2 = k & (LEAF_LENGTH - 1);
		ASSERT(i1 < ROOT_LENGTH);
		root_[i1]->values[i2] = v;
	}
	bool Ensure(Number start, size_t n) {
		for (Number key = start; key <= start + n - 1;) {
			const Number i1 = key >> LEAF_BITS;
			// Check for overflow
			if (i1 >= ROOT_LENGTH)
				return false;
			// Make 2nd level node if necessary
			if (root_[i1] == NULL) {
				//Leaf* leaf = reinterpret_cast<Leaf*>((*allocator_)(sizeof(Leaf)));
				//if (leaf == NULL) return false;
				static ObjectPool<Leaf> leafpool;
				//Leaf* leaf = reinterpret_cast<Leaf*>((*allocator_)
				Leaf* leaf = (Leaf*)leafpool.New();
				memset(leaf, 0, sizeof(*leaf));
				root_[i1] = leaf;
			}
			// Advance key past whatever is covered by this leaf node
			key = ((key >> LEAF_BITS) + 1) << LEAF_BITS;
		}
		return true;
	}
	void PreallocateMoreMemory() {
		// Allocate enough to keep track of all possible pages
		
			Ensure(0, 1 << BITS);
	}
};
// Three-level radix tree
template <int BITS>
class TCMalloc_PageMap3 {
private:
	// How many bits should we consume at each interior level
	static const int INTERIOR_BITS = (BITS + 2) / 3; // Round-up
	static const int INTERIOR_LENGTH = 1 << INTERIOR_BITS;
	// How many bits should we consume at leaf level
	static const int LEAF_BITS = BITS - 2 * INTERIOR_BITS;
	static const int LEAF_LENGTH = 1 << LEAF_BITS;
	// Interior node
	struct Node {
		Node* ptrs[INTERIOR_LENGTH];
	};
	// Leaf node
	struct Leaf {
		void* values[LEAF_LENGTH];
	};
	Node* root_;                          // Root of radix tree
	void* (*allocator_)(size_t);          // Memory allocator
	Node* NewNode() {
		Node* result = reinterpret_cast<Node*>((*allocator_)(sizeof(Node)));
		if (result != NULL) {
			memset(result, 0, sizeof(*result));
		}
		return result;
	}
public:
	typedef uintptr_t Number;
	explicit TCMalloc_PageMap3(void* (*allocator)(size_t)) {
		allocator_ = allocator;
		root_ = NewNode();
	}
	void* get(Number k) const {
		const Number i1 = k >> (LEAF_BITS + INTERIOR_BITS);
		const Number i2 = (k >> LEAF_BITS) & (INTERIOR_LENGTH - 1);
		const Number i3 = k & (LEAF_LENGTH - 1);
		if ((k >> BITS) > 0 ||
			root_->ptrs[i1] == NULL || root_->ptrs[i1]->ptrs[i2] == NULL) {
			return NULL;
		}
		return reinterpret_cast<Leaf*>(root_->ptrs[i1]->ptrs[i2])->values[i3];
	}
	void set(Number k, void* v) {
		ASSERT(k >> BITS == 0);
		const Number i1 = k >> (LEAF_BITS + INTERIOR_BITS);
		const Number i2 = (k >> LEAF_BITS) & (INTERIOR_LENGTH - 1);
		const Number i3 = k & (LEAF_LENGTH - 1);
		reinterpret_cast<Leaf*>(root_->ptrs[i1]->ptrs[i2])->values[i3] = v;
	}
	bool Ensure(Number start, size_t n) {
		for (Number key = start; key <= start + n - 1;) {
			const Number i1 = key >> (LEAF_BITS + INTERIOR_BITS);
			const Number i2 = (key >> LEAF_BITS) & (INTERIOR_LENGTH - 1);
			// Check for overflow
			if (i1 >= INTERIOR_LENGTH || i2 >= INTERIOR_LENGTH)
				return false;
			// Make 2nd level node if necessary
			if (root_->ptrs[i1] == NULL) {
				Node* n = NewNode();
				if (n == NULL) return false;
				root_->ptrs[i1] = n;
			}
			// Make leaf node if necessary
			if (root_->ptrs[i1]->ptrs[i2] == NULL) {
				Leaf* leaf = reinterpret_cast<Leaf*>((*allocator_)
					(sizeof(Leaf)));
				if (leaf == NULL) return false;
				memset(leaf, 0, sizeof(*leaf));
				root_->ptrs[i1]->ptrs[i2] = reinterpret_cast<Node*>(leaf);
			}
			// Advance key past whatever is covered by this leaf node
			key = ((key >> LEAF_BITS) + 1) << LEAF_BITS;
		}
		return true;
	}
	void PreallocateMoreMemory() {
	}
};

PageCache.h

#pragma once
#include "Common.h"
#include "ObjectPool.h"
template <int BITS>
class TCMalloc_PageMap1 {
private:
	static const int LENGTH = 1 << BITS;
	void** array_;
public:
	typedef uintptr_t Number;
	explicit TCMalloc_PageMap1(void* (*allocator)(size_t)) {
		//array_ = reinterpret_cast<void**>((*allocator)(sizeof(void*) << BITS));
		size_t size = sizeof(void*) << BITS;
		size_t  alignSize = SizeClass::_RoundUp(size,1 << PAGE_SHIFT);

		array_ = SystemAlloc(alignSize >> PAGE_SHIFT);
		memset(array_, 0, sizeof(void*) << BITS);
	}
	// Return the current value for KEY. Returns NULL if not yet set,
	// or if k is out of range.
	void* get(Number k) const {
		if ((k >> BITS) > 0) {
			return NULL;
		}
		return array_[k];
	}
	// REQUIRES "k" is in range "[0,2^BITS-1]".
	// REQUIRES "k" has been ensured before.
	//
	// Sets the value 'v' for key 'k'.
	void set(Number k, void* v) {
		array_[k] = v;
	}
};
// Two-level radix tree
template <int BITS>
class TCMalloc_PageMap2 {
private:
	// Put 32 entries in the root and (2^BITS)/32 entries in each leaf.
	static const int ROOT_BITS = 5;
	static const int ROOT_LENGTH = 1 << ROOT_BITS;
	static const int LEAF_BITS = BITS - ROOT_BITS;
	static const int LEAF_LENGTH = 1 << LEAF_BITS;
	比特就业课
		// Leaf node
		struct Leaf {
		void* values[LEAF_LENGTH];
	};
	Leaf* root_[ROOT_LENGTH];             // Pointers to 32 child nodes
	void* (*allocator_)(size_t);          // Memory allocator
public:
	typedef uintptr_t Number;
	//explicit TCMalloc_PageMap2(void* (*allocator)(size_t)) 
	explicit TCMalloc_PageMap2() {
		//allocator_ = allocator;
		memset(root_, 0, sizeof(root_));
	}
	void* get(Number k) const {
		const Number i1 = k >> LEAF_BITS;
		const Number i2 = k & (LEAF_LENGTH - 1);
		if ((k >> BITS) > 0 || root_[i1] == NULL) {
			return NULL;
		}
		return root_[i1]->values[i2];
	}
	void set(Number k, void* v) {
		const Number i1 = k >> LEAF_BITS;
		const Number i2 = k & (LEAF_LENGTH - 1);
		ASSERT(i1 < ROOT_LENGTH);
		root_[i1]->values[i2] = v;
	}
	bool Ensure(Number start, size_t n) {
		for (Number key = start; key <= start + n - 1;) {
			const Number i1 = key >> LEAF_BITS;
			// Check for overflow
			if (i1 >= ROOT_LENGTH)
				return false;
			// Make 2nd level node if necessary
			if (root_[i1] == NULL) {
				//Leaf* leaf = reinterpret_cast<Leaf*>((*allocator_)(sizeof(Leaf)));
				//if (leaf == NULL) return false;
				static ObjectPool<Leaf> leafpool;
				//Leaf* leaf = reinterpret_cast<Leaf*>((*allocator_)
				Leaf* leaf = (Leaf*)leafpool.New();
				memset(leaf, 0, sizeof(*leaf));
				root_[i1] = leaf;
			}
			// Advance key past whatever is covered by this leaf node
			key = ((key >> LEAF_BITS) + 1) << LEAF_BITS;
		}
		return true;
	}
	void PreallocateMoreMemory() {
		// Allocate enough to keep track of all possible pages
		
			Ensure(0, 1 << BITS);
	}
};
// Three-level radix tree
template <int BITS>
class TCMalloc_PageMap3 {
private:
	// How many bits should we consume at each interior level
	static const int INTERIOR_BITS = (BITS + 2) / 3; // Round-up
	static const int INTERIOR_LENGTH = 1 << INTERIOR_BITS;
	// How many bits should we consume at leaf level
	static const int LEAF_BITS = BITS - 2 * INTERIOR_BITS;
	static const int LEAF_LENGTH = 1 << LEAF_BITS;
	// Interior node
	struct Node {
		Node* ptrs[INTERIOR_LENGTH];
	};
	// Leaf node
	struct Leaf {
		void* values[LEAF_LENGTH];
	};
	Node* root_;                          // Root of radix tree
	void* (*allocator_)(size_t);          // Memory allocator
	Node* NewNode() {
		Node* result = reinterpret_cast<Node*>((*allocator_)(sizeof(Node)));
		if (result != NULL) {
			memset(result, 0, sizeof(*result));
		}
		return result;
	}
public:
	typedef uintptr_t Number;
	explicit TCMalloc_PageMap3(void* (*allocator)(size_t)) {
		allocator_ = allocator;
		root_ = NewNode();
	}
	void* get(Number k) const {
		const Number i1 = k >> (LEAF_BITS + INTERIOR_BITS);
		const Number i2 = (k >> LEAF_BITS) & (INTERIOR_LENGTH - 1);
		const Number i3 = k & (LEAF_LENGTH - 1);
		if ((k >> BITS) > 0 ||
			root_->ptrs[i1] == NULL || root_->ptrs[i1]->ptrs[i2] == NULL) {
			return NULL;
		}
		return reinterpret_cast<Leaf*>(root_->ptrs[i1]->ptrs[i2])->values[i3];
	}
	void set(Number k, void* v) {
		ASSERT(k >> BITS == 0);
		const Number i1 = k >> (LEAF_BITS + INTERIOR_BITS);
		const Number i2 = (k >> LEAF_BITS) & (INTERIOR_LENGTH - 1);
		const Number i3 = k & (LEAF_LENGTH - 1);
		reinterpret_cast<Leaf*>(root_->ptrs[i1]->ptrs[i2])->values[i3] = v;
	}
	bool Ensure(Number start, size_t n) {
		for (Number key = start; key <= start + n - 1;) {
			const Number i1 = key >> (LEAF_BITS + INTERIOR_BITS);
			const Number i2 = (key >> LEAF_BITS) & (INTERIOR_LENGTH - 1);
			// Check for overflow
			if (i1 >= INTERIOR_LENGTH || i2 >= INTERIOR_LENGTH)
				return false;
			// Make 2nd level node if necessary
			if (root_->ptrs[i1] == NULL) {
				Node* n = NewNode();
				if (n == NULL) return false;
				root_->ptrs[i1] = n;
			}
			// Make leaf node if necessary
			if (root_->ptrs[i1]->ptrs[i2] == NULL) {
				Leaf* leaf = reinterpret_cast<Leaf*>((*allocator_)
					(sizeof(Leaf)));
				if (leaf == NULL) return false;
				memset(leaf, 0, sizeof(*leaf));
				root_->ptrs[i1]->ptrs[i2] = reinterpret_cast<Node*>(leaf);
			}
			// Advance key past whatever is covered by this leaf node
			key = ((key >> LEAF_BITS) + 1) << LEAF_BITS;
		}
		return true;
	}
	void PreallocateMoreMemory() {
	}
};

PageCache.cpp

#include "PageCache.h"
PageCache PageCache::_sInst;

Span* PageCache::NewSpan(size_t k)//获取k页的span
{
	//eg:只有一个128页的，需要两页的-》128分为2span和126span，2返回给central cache,126挂道对应的桶上
	//如果central cache中的span usecount=0,说明切分给thread cache小块内存都还回来了，
	//则central cache把span还给page cache,page cache通过页号查看相邻页是否空闲，是就合并出更大的page,解决内存碎片问题
	assert(k > 0 && k < NPAGES);
	
	if (k > NPAGES - 1)
	{
		void* ptr = SystemAlloc(k);//大于最大页数直接找堆要
		//Span* span = new Span;
		Span* span = spanPool.New();
		span->_pageId = (PAGE_ID)ptr >> PAGE_SHIFT;
		span->n = k;
		//_idSpanMap[span->_pageId] = span;
		_idSpanMap.set(span->_pageId,span);
		return span;
	}
	
	if (!_spanList[k].Empty())//第k个桶里面有没有span
	{
		
		Span* KSpan = _spanList[k].PopFront();
		
		//建立id和span的映射关系方便centralcache回收小块内存时查找对应的span
		for (PAGE_ID i = 0; i < KSpan->n; i++)
		{
			//_idSpanMap[KSpan->_pageId + i] = KSpan;
			_idSpanMap.set(KSpan->_pageId + i, KSpan);
		}
		
		return KSpan;
	}

	//第k个桶里是空的,检测后面的桶里有没有span,如果有进行切分
	//切分成一个k页的span和一个 n-k 页的span
	//k页的span返回给central cache,n-k 页的span挂到第 n-k 号桶中去
	for (size_t i = k ; i < NPAGES; i++)
	{		
		if (!_spanList[i].Empty())
		{
			Span* nSpan = _spanList[i].PopFront();
			//Span* KSpan = new Span;
			Span* KSpan = spanPool.New();
			

			//在nSpan的头部切下K页
			//k页span返回，nSpan再挂到对应映射
			KSpan->_pageId = nSpan->_pageId;
			KSpan->n = k;//kSpan页数变为k
			nSpan->_pageId += k;//nSpan 页号变为 += k
			nSpan->n -= k;
			
			_spanList[nSpan->n].PushFront(nSpan);//把剩余的页挂到对应的位置

			
			//存储nSpan的首尾页号跟span映射,方便page cache回收内存时进行合并查找
			//_idSpanMap[nSpan->_pageId] = nSpan;
			//_idSpanMap[nSpan->_pageId + nSpan->n - 1] = nSpan
			_idSpanMap.set(nSpan->_pageId, nSpan);
			_idSpanMap.set(nSpan->_pageId + nSpan->n - 1, nSpan);
			

			//建立id和span的映射，方便central cache回收查找对应的span
			for (PAGE_ID i = 0; i < KSpan->n; i++)
			{
				//_idSpanMap[KSpan->_pageId + i] = KSpan;
				_idSpanMap.set(KSpan->_pageId + i, KSpan);
			}
			
			
			return KSpan;
		}
	}

	//没有大页span
	//找堆要128页的span
	Span* bigSpan = spanPool.New();
	//Span* bigSpan = new Span;
	void* ptr = SystemAlloc(NPAGES - 1);
	bigSpan->_pageId = (PAGE_ID)ptr >> PAGE_SHIFT;
	bigSpan->n = NPAGES - 1;
	_spanList[bigSpan->n].PushFront(bigSpan);
	return NewSpan(k);

}

Span* PageCache::MapObjectToSpan(void* obj)
{
	
	PAGE_ID id = ((PAGE_ID)obj >> PAGE_SHIFT);//找出页号


	
	auto ret = (Span*)_idSpanMap.get(id);
	assert(ret != nullptr);
	return ret;
}
void PageCache::ReleaseSpanToPageCache(Span* span)
{
	//对span前后的页进行合并，解决内存碎片问题

	if (span->n > NPAGES - 1)
	{
		//大于128页的直接还给堆
		void* ptr = (void*)(span->_pageId << PAGE_SHIFT);
		SystemFree(ptr);
		//delete span;
		spanPool.Delete(span);
		return;
	}
	//向前合并
	while (1)
	{
		PAGE_ID prevId = span->_pageId - 1;
	
		auto ret = (Span*)_idSpanMap.get(prevId);
		if (ret == nullptr)
		{
			break;
		}
		
		Span* prevspan = ret;
		if (prevspan->_isUse == true)
		{//前面相邻页的span在使用
			break;
		}
		if (prevspan->n + span->n > NPAGES - 1)
		{
			//合并数超过128，没办法管理
			break;
		}
		//合并
		span->_pageId = prevspan->_pageId;
		span->n += prevspan->n;
		_spanList[prevspan->n].Erase(prevspan);
		//delete prevspan;
		
		spanPool.Delete(prevspan);
	}

	//向后合并
	while (1)
	{
		PAGE_ID nextId = span->_pageId + span->n;
		
		auto ret = (Span*)_idSpanMap.get(nextId);
		if (ret == nullptr)
		{
			break;
		}
		Span* nextspan = ret;
		if (nextspan->_isUse == true)
		{
			break;
		}
		if (nextspan->n + span->n > NPAGES - 1)
		{
			break;
		}
		span->n += nextspan->n;
		_spanList[nextspan->n].Erase(nextspan);
		//delete(nextspan);
		spanPool.Delete(nextspan);

	}
	//前后都合并过了

	_spanList[span->n].PushFront(span);
	span->_isUse = false;

	//方便其他把此span合并
	//_idSpanMap[span->_pageId] = span;
	//_idSpanMap[span->_pageId + span->n - 1] = span;

	_idSpanMap.set(span->_pageId, span);
	_idSpanMap.set(span->_pageId + span->n - 1, span);

}
//只有Span* NewSpan(size_t k)   void ReleaseSpanToPageCache(Span* span)
//这两个函数中去建立id和span的映射(回去写)
//基数树，写之前会提前开好空间，写数据过程中，不会动数据结构
//读写是分离的。线程1对一个位置读写的时候，线程2不可以对这个位置读写

转载至：https://zhuanlan.zhihu.com/p/582514123