C++0X（C++11）的std:mutex、std::thread与Windows API性能对比

转载请注明原创出处：http://aigo.iteye.com/blog/1908084

C++0X标准提供的std::mutex和std::thread两个接口开发多线程同步的应用非常方便，而且可以跨平台，自己做了一下测试，发现这个跨平台的代价还是很大的，我分别用std::mutex与Windows的CRITICAL_SECTION、std::thead和WIndows的CreateThread接口做了对比，测试代码如下：

#include "stdafx.h"
#include <mutex>
#include <atomic> 
#include <iostream>
#include <time.h>
#include <thread>
#include <list>
#include <atomic> 
#include <Windows.h>

using namespace std;

#define MAX_THREADS 16

// 全局的结果数据 
long total1 = 0; 
long total2 = 0;
std::atomic<long> total;

std::mutex m_lock;

CRITICAL_SECTION m_Lock2;

void use_std_mutex();
void use_win_critical();
void use_win_thread();


void test_mutex()
{
	for(int i=0; i<1000000;++i)
	{
		m_lock.lock();
		total1 += 1;     
		m_lock.unlock();
	}
}

void test_critical()
{
	for(int i=0; i<1000000;++i)
	{
		EnterCriticalSection(&m_Lock2);
		total2 += 1;     
		LeaveCriticalSection(&m_Lock2);
	}
}

int main(int argc, char* argv[])
{
	use_std_mutex();
	use_win_critical();
	use_win_thread();
	
	return 0;
}

void use_std_mutex()
{
	std::list<std::thread*> threadlist;

	//测试mutex
	printf("testing mutex...\n");
	clock_t start = clock();
	for(int i=0; i<MAX_THREADS; ++i) 
	{
		std::thread *t1 = new std::thread((&test_mutex));
		threadlist.push_back(t1);
	}
	for(std::list<std::thread*>::const_iterator i = threadlist.begin(); i != threadlist.end(); i++ )
	{
		(*i)->join();
	}
	clock_t finish = clock();
	printf("result:%d\n", total1);
	printf("cost:%dms\n", finish - start);
	for(std::list<std::thread*>::const_iterator i = threadlist.begin(); i != threadlist.end(); i++ )
	{
		delete(*i);
	}
}

void use_win_critical()
{
	//测试Critical
	InitializeCriticalSection(&m_Lock2);
	std::list<std::thread*> threadlist;
	printf("testing critical...\n");
	clock_t start = clock();
	for(int i=0; i<MAX_THREADS; ++i) 
	{
		std::thread *t1 = new std::thread((&test_critical));
		threadlist.push_back(t1);
	}
	for(std::list<std::thread*>::const_iterator i = threadlist.begin(); i != threadlist.end(); i++ )
	{
		(*i)->join();
	}
	clock_t finish = clock();
	printf("result:%d\n", total2);
	printf("cost:%dms\n", finish - start);
	for(std::list<std::thread*>::const_iterator i = threadlist.begin(); i != threadlist.end(); i++ )
	{
		delete(*i);
	}
}


#define BUF_SIZE 255
long total3 = 0; 
CRITICAL_SECTION m_Lock3;

DWORD WINAPI MyThreadFunction( LPVOID lpParam );
//使用Windows线程测试
void use_win_thread()
{
	DWORD   dwThreadIdArray[MAX_THREADS];
	HANDLE  hThreadArray[MAX_THREADS]; 

	InitializeCriticalSection(&m_Lock3);
	printf("testing use_win_thread...\n");

	clock_t start = clock();
	for( int i=0; i<MAX_THREADS; i++ )
	{
		hThreadArray[i] = CreateThread( 
			NULL,                   
			0,                      
			MyThreadFunction,       
			NULL,					
			0,                      
			&dwThreadIdArray[i]);   

	} 

	WaitForMultipleObjects(MAX_THREADS, hThreadArray, TRUE, INFINITE);

	clock_t finish = clock();
	printf("result:%d\n", total3);
	printf("cost:%dms\n", finish - start);

	for(int i=0; i<MAX_THREADS; i++)
	{
		CloseHandle(hThreadArray[i]);
	}

}

DWORD WINAPI MyThreadFunction( LPVOID lpParam ) 
{ 
	for(int i=0; i<1000000;++i)
	{
		EnterCriticalSection(&m_Lock3);
		total3 += 1;     
		LeaveCriticalSection(&m_Lock3);
	}
	return 0; 
} 

 

 

 

测试环境：
硬件：i7 2630qm 4核

系统：Windows7 64bit 旗舰版  SP1

程序：VS2012 Release win32

 

测试结果：

2线程抢占：

testing mutex...

result:2000000

cost:628ms

testing critical...

result:2000000

cost:132ms

testing use_win_thread...

result:2000000

cost:98ms

 

4线程抢占：

testing mutex...

result:4000000

cost:1150ms

testing critical...

result:4000000

cost:266ms

testing use_win_thread...

result:4000000

cost:216ms

 

8线程抢占：

testing mutex...

result:8000000

cost:2855ms

testing critical...

result:8000000

cost:582ms

testing use_win_thread...

result:8000000

cost:461ms

 

16线程抢占：

testing mutex...

result:16000000

cost:138052ms

testing critical...

result:16000000

cost:1448ms

testing use_win_thread...

result:16000000

cost:1169ms

 

结论：如果想追求高性能，C++11的std::mutex不要用，std::thread性能损耗不大，要用随喜。