linux下mutex与atomic性能比较

一种是用boost::atomic；一种直接加锁；代码很简单：

#include <boost/atomic/atomic.hpp>
#include <iostream>
#include <stdlib.h>
#include <boost/thread/mutex.hpp>
#include <pthread.h>
#include <stdio.h> 
#include <sys/time.h> 
#include <time.h> 

static int loop_num = 10000;
boost::atomic<int> a(0);
static boost::mutex mtx_;
static void* test_atomic(void* agr){
	int num = *(int*)agr;
	struct timeval t_start,t_end; 

	long cost_time = 0; 
	gettimeofday(&t_start, NULL); 
	long start = ((long)t_start.tv_sec)*1000+(long)t_start.tv_usec/1000; 

	while (num--){
		++a;
	}

	gettimeofday(&t_end, NULL); 
	long end = ((long)t_end.tv_sec)*1000+(long)t_end.tv_usec/1000; 
	cost_time = end - start; 
	std::cout << "loopnum:" << loop_num << " test_atomic cost "<< cost_time << " ms\n";
	return NULL;
}

static int b = 0;
static void* test_lock(void* agr){
	int num = *(int*)agr;

	struct timeval t_start,t_end; 

	long cost_time = 0; 
	gettimeofday(&t_start, NULL); 
	long start = ((long)t_start.tv_sec)*1000+(long)t_start.tv_usec/1000; 

	while(num--){
		{
			boost::mutex::scoped_lock lock(mtx_);
			++b;
		}
	}
	gettimeofday(&t_end, NULL); 
	long end = ((long)t_end.tv_sec)*1000+(long)t_end.tv_usec/1000; 
	cost_time = end - start; 
	std::cout  << "loopnum:" << loop_num << "test_lock cost "<< cost_time << " ms\n";
	return NULL;
}
int main(int agrc, char** argv){
	if (agrc < 2){
		std::cout << "please input num:" << std::endl;
		return 0;
	}
	pthread_t main_tid;
	int num1 = atoi((char*)argv[1]);
	int num2 = atoi((char*)argv[2]);
	if (num1 == 0){
		pthread_create(&main_tid, NULL, &test_atomic, &num2); //创建线程
    		pthread_join(main_tid, NULL);
	}
	else{
		pthread_create(&main_tid, NULL, &test_lock, &num2); //创建线程
    		pthread_join(main_tid, NULL);
	}
}

allen@allen-ThinkPad-Edge-E431:~/code$ ./test 0 1000
loopnum:10000 test_atomic cost 0 ms
allen@allen-ThinkPad-Edge-E431:~/code$ ./test 1 1000
loopnum:10000test_lock cost 0 ms
allen@allen-ThinkPad-Edge-E431:~/code$ ./test 1 100000
loopnum:10000test_lock cost 9 ms
allen@allen-ThinkPad-Edge-E431:~/code$ ./test 0 100000
loopnum:10000 test_atomic cost 2 ms

随着loop数目的增大差距越来越大；与两者实现有关,一种是临界区阻塞，一种用了compare change技术，甚至看内部实现用了汇编。

在我的笔记本结果是这样，也说明了问题。