C++_项目高并发内存池——优化-测试与查错Span链表中自由链表尾未置空--与malloc在多线程的情况下对比性能

测试代码：

#include"ConcurrentAlloc.h"

//ntimes：一轮申请与释放内存的此时 rounds：一共测试几轮 nworks线程数

void TestBenchMarkMalloc(size_t ntimes, size_t rounds, size_t nworks) {
	std::cout << "Malloc Test" << std::endl;
	std::vector<std::thread>ThreadPool(nworks);

	//计算每个线程申请释放的时间再相加，使用的是原子性的相加，C++11语法
	std::atomic<size_t>malloc_time = 0;
	std::atomic<size_t>free_time = 0;

	for (size_t i = 0; i < nworks; i++) {
		ThreadPool[i] = std::thread(
			[&, i]() {//表示引用传递捕捉所有父作用域中的变量(包括this)
				std::vector<void*>Pool;//保存申请的空间地址
				Pool.reserve(ntimes);
				for (size_t j = 0; j < rounds; j++) {
					size_t begin = clock();
					for (size_t k = 0; k < ntimes; k++) {
						Pool.push_back(malloc(8 + k));
					}
					size_t end = clock();

					size_t begin2 = clock();
					for (size_t k = 0; k < ntimes; k++) {
						free(Pool[k]);
					}
					size_t end2 = clock();
					Pool.clear();
					malloc_time += (end - begin);
					free_time += (end2 - begin2);
				}
			}
		);//C++11 lambda表达式
	}

	for (auto& e : ThreadPool) {
		e.join();
	}
	std::cout << nworks << "个线程 单个线程测试"<<rounds<<"轮 每轮" << ntimes << "次申请与释放内存" << std::endl << 
		"申请时间" << malloc_time.load() << "ms 释放时间" << free_time.load() << "ms\n总时间" << free_time.load() + malloc_time.load() <<"ms" << std::endl;
	std::cout << "########################################" << std::endl;

}

void TestConcurrentAlloc(size_t ntimes, size_t rounds, size_t nworks) {
	std::cout << "ConcurrentAlloc Test" << std::endl;
	std::vector<std::thread>ThreadPool(nworks);
	std::atomic<size_t>Alloc_time = 0;
	std::atomic<size_t>Delete_time = 0; 

	for (size_t i = 0; i < nworks; i++) {
		ThreadPool[i] = std::thread(
			[&, i]() {
				std::vector<void*>Pool;
				Pool.reserve(ntimes);
				for (size_t j = 0; j < rounds; j++) {
					size_t begin = clock();
					for (size_t k = 0; k < ntimes; k++) {
						Pool.push_back(ConcurrentAlloc(8 ));
					}
					size_t end = clock();

					size_t begin2 = clock();
					for (size_t k = 0; k < ntimes; k++) {
						ConcurrentFree(Pool[k]);
					}
					size_t end2 = clock();
					Pool.clear();
					Alloc_time += (end - begin);
					Delete_time += (end2 - begin2);
				}
			}
		);//C++11 lambda表达式
	}

	for (auto& e : ThreadPool) {
		e.join();
	}
	std::cout << nworks << "个线程 单个线程测试" << rounds << "轮 每轮" << ntimes << "次申请与释放内存" << std::endl <<
		"申请时间" << Alloc_time.load() << "ms 释放时间" << Delete_time.load() << "ms\n总时间" << Alloc_time.load() + Delete_time.load() << "ms" << std::endl;
	std::cout << "########################################" << std::endl;

}

int main() {
	TestBenchMarkMalloc(10000,4,10);
	TestConcurrentAlloc(10000, 4, 10);
	return 0;
}

上述代码分别测试了在多线程下malloc和包并发内存池下申请与释放所需要的时间。

注意：
在测试时发现的错误：
1. 在切分Span时，Span中自由链表尾没有置空导致切分空间出错，需要加上NextObj(tail) = nullptr

2.当前桶中存在Span时直接返回了，没有将Span与页号建立起映射

//看当前位置桶中是否有Span
if (!_SpanList[NumPage].Empty()) {
	Span*NumPageSpan =_SpanList[NumPage].Pop();
	for (PAGE_ID i = 0; i < NumPageSpan->_Num; i++) {//切了Num页
		IdSpanMap[NumPageSpan->_PageID + i] = NumPageSpan;
	}
	return NumPageSpan;
}

Span* CentralCache::GetSpan(SpanList& List, size_t size) {
	//在哈希桶对应位置Span链表中找是否有Span，没有就向PageCache申请空间

	//遍历桶的Span链表
	Span* it = List.begin();
	while (it != List.end()) {
		if (it->FreeList != nullptr) {
			return it;//这个Span有空间
		}
		else {
			//Span没有空间，继续找下一个链表Span
			it = it->_next;
		}
	}
	//先把CentralCache的桶锁解开，如果其他线程释放内存不会阻塞
	List._mtx.unlock();
	//没有空闲的Span只能找PageCache,需要加锁，PageCache只能由一个线程访问
	//size是单个对象的大小
	PageCache::GetInst()->_PageMtx.lock();
	Span* span=PageCache::GetInst()->NewSpan(SizeClass::NumForPage(size));
	span->IsUse = true;
	span->ObjectSize = size;//保存size，为了让ConcurrentFree 释放空间时不需要传大小
	PageCache::GetInst()->_PageMtx.unlock();

	//获得了一块大Span，这块Span这时被线程单独看到，不需要加锁（没有挂到桶上）
	//Span起始地址
	char* start = (char*)((span->_PageID) << PAGESIZE);
	size_t ByteSize = (span->_Num) << PAGESIZE;
	char* end = start + ByteSize;
	//把Span内部大块内存切成自由链表链接起来
	span->FreeList = start;
	start += size;//自由链表的头节点
	void* tail = span->FreeList;
	while (start < end) {
		NextObj(tail) = start;
		tail = NextObj(tail);
		start += size;
	}
	NextObj(tail) = nullptr;//出现BUG的原因
	List._mtx.lock();
	List.Insert(List.begin(), span);//将Span挂到桶上，此时需要加桶锁
	return span;
}

测试截图：

如上图可以看出高并发内存池在多线程下申请空间的效率比malloc要高很多

修改后，并测试代码位置：

Github
Gitee