C++內存池的簡單原理及實現

爲什麼要用內存池

C++程序默認的內存管理（new，delete，malloc，free）會頻繁地在堆上分配和釋放內存，導致性能的損失，產生大量的內存碎片，降低內存的利用率。默認的內存管理因爲被設計的比較通用，所以在性能上並不能做到極致。
因此，很多時候需要根據業務需求設計專用內存管理器，便於針對特定數據結構和使用場合的內存管理，比如：內存池。

內存池原理

內存池的思想是，在真正使用內存之前，預先申請分配一定數量、大小預設的內存塊留作備用。當有新的內存需求時，就從內存池中分出一部分內存塊，若內存塊不夠再繼續申請新的內存，當內存釋放後就回歸到內存塊留作後續的複用，使得內存使用效率得到提升，一般也不會產生不可控制的內存碎片。

內存池設計

算法原理：

1. 預申請一個內存區chunk，將內存中按照對象大小劃分成多個內存塊block
2. 維持一個空閒內存塊鏈表，通過指針相連，標記頭指針爲第一個空閒塊
3. 每次新申請一個對象的空間，則將該內存塊從空閒鏈表中去除，更新空閒鏈表頭指針
4. 每次釋放一個對象的空間，則重新將該內存塊加到空閒鏈表頭
5. 如果一個內存區佔滿了，則新開闢一個內存區，維持一個內存區的鏈表，同指針相連，頭指針指向最新的內存區，新的內存塊從該區內重新劃分和申請

如圖所示：

內存池實現

memory_pool.hpp

#ifndef _MEMORY_POOL_H_
#define _MEMORY_POOL_H_

#include <stdint.h>
#include <mutex>

template<size_t BlockSize, size_t BlockNum = 10>
class MemoryPool
{
public:
	MemoryPool()
	{
		std::lock_guard<std::mutex> lk(mtx); // avoid race condition

		// init empty memory pointer
		free_block_head = NULL;
		mem_chunk_head = NULL;
	}

	~MemoryPool()
	{
		std::lock_guard<std::mutex> lk(mtx); // avoid race condition

		// destruct automatically
		MemChunk* p;
		while (mem_chunk_head)
		{
			p = mem_chunk_head->next;
			delete mem_chunk_head;
			mem_chunk_head = p;
		}
	}

	void* allocate()
	{
		std::lock_guard<std::mutex> lk(mtx); // avoid race condition

		// allocate one object memory

		// if no free block in current chunk, should create new chunk
		if (!free_block_head)
		{
			// malloc mem chunk
			MemChunk* new_chunk = new MemChunk;
			new_chunk->next = NULL;

			// set this chunk's first block as free block head
			free_block_head = &(new_chunk->blocks[0]);

			// link the new chunk's all blocks
			for (int i = 1; i < BlockNum; i++)
				new_chunk->blocks[i - 1].next = &(new_chunk->blocks[i]);
			new_chunk->blocks[BlockNum - 1].next = NULL; // final block next is NULL
			
			if (!mem_chunk_head)
				mem_chunk_head = new_chunk;
			else
			{
				// add new chunk to chunk list
				mem_chunk_head->next = new_chunk;
				mem_chunk_head = new_chunk;
			}
		}

		// allocate the current free block to the object
		void* object_block = free_block_head;
		free_block_head = free_block_head->next; 

		return object_block;
	}

	void* allocate(size_t size)
	{
		std::lock_guard<std::mutex> lk(array_mtx); // avoid race condition for continuous memory

		// calculate objects num
		int n = size / BlockSize;

		// allocate n objects in continuous memory
		
		// FIXME: make sure n > 0
		void* p = allocate();

		for (int i = 1; i < n; i++)
			allocate();

		return p;
	}

	void deallocate(void* p)
	{
		std::lock_guard<std::mutex> lk(mtx); // avoid race condition

		// free object memory
		FreeBlock* block = static_cast<FreeBlock*>(p);
		block->next = free_block_head; // insert the free block to head
		free_block_head = block;
	}

private:
	// free node block, every block size exactly can contain one object
	struct FreeBlock
	{
		unsigned char data[BlockSize];
		FreeBlock* next;
	};

	FreeBlock* free_block_head;

	// memory chunk, every chunk contains blocks number with fixed BlockNum
	struct MemChunk
	{
		FreeBlock blocks[BlockNum];
		MemChunk* next;
	};

	MemChunk* mem_chunk_head;

	// thread safe related
	std::mutex mtx;
	std::mutex array_mtx;
};

#endif // !_MEMORY_POOL_H_

main.cpp

#include <iostream>
#include "memory_pool.hpp"

class MyObject
{
public:
	MyObject(int x): data(x)
	{
		//std::cout << "contruct object" << std::endl;
	}

	~MyObject()
	{
		//std::cout << "destruct object" << std::endl;
	}

	int data;

	// override new and delete to use memory pool
	void* operator new(size_t size);
	void operator delete(void* p);
	void* operator new[](size_t size);
	void operator delete[](void* p);
};

// define memory pool with block size as class size
MemoryPool<sizeof(MyObject), 3> gMemPool;


void* MyObject::operator new(size_t size)
{
	//std::cout << "new object space" << std::endl;
	return gMemPool.allocate();
}

void MyObject::operator delete(void* p)
{
	//std::cout << "free object space" << std::endl;
	gMemPool.deallocate(p);
}

void* MyObject::operator new[](size_t size)
{
	// TODO: not supported continuous memoery pool for now
	//return gMemPool.allocate(size);
	return NULL;
}
void MyObject::operator delete[](void* p)
{
	// TODO: not supported continuous memoery pool for now
	//gMemPool.deallocate(p);
}

int main(int argc, char* argv[])
{
	MyObject* p1 = new MyObject(1);
	std::cout << "p1 " << p1 << " " << p1->data<< std::endl;

	MyObject* p2 = new MyObject(2);
	std::cout << "p2 " << p2 << " " << p2->data << std::endl;
	delete p2;

	MyObject* p3 = new MyObject(3);
	std::cout << "p3 " << p3 << " " << p3->data << std::endl;

	MyObject* p4 = new MyObject(4);
	std::cout << "p4 " << p4 << " " << p4->data << std::endl;

	MyObject* p5 = new MyObject(5);
	std::cout << "p5 " << p5 << " " << p5->data << std::endl;

	MyObject* p6 = new MyObject(6);
	std::cout << "p6 " << p6 << " " << p6->data << std::endl;

	delete p1;
	delete p2;
	//delete p3;
	delete p4;
	delete p5;
	delete p6;

	getchar();
	return 0;
}

運行結果

p1 00000174BEDE0440 1
p2 00000174BEDE0450 2
p3 00000174BEDE0450 3
p4 00000174BEDE0460 4
p5 00000174BEDD5310 5
p6 00000174BEDD5320 6

可以看到內存地址是連續，並且回收一個節點後，依然有序地開闢內存
對象先開闢內存再構造，先析構再釋放內存

注意

• 在內存分配和釋放的環節需要加鎖來保證線程安全
• 還沒有實現對象數組的分配和釋放