CSAPP学习笔记（三）多级存储和优化的思考

迈不过友情╰ 2024-04-17 20:08 56阅读 0赞

### 1. 简介 ###

现代计算从CPU开始到主存甚至云端，其实是一个多级存储的金字塔结构，越往上速度越快容量越小，每一级均是上一级的缓存，而这也导致了写代码时不同位置的存储/读取对代码性能会产生巨大的影响。

### 2. 示例 ###

这里我写了一个小测试用以验证我的想法是否正确。如下代码中， test1-test4要做的事累加，并将和存放在类成员m\_val中。使用的方法分别是：

*  test1()采用局部变量保存累加和，最后返回并赋值给m\_val
 *  test2()直接在m\_val中累加
 *  test3()使用全局变量保存
 *  test4()使用全局静态变量保存

#include <iostream>
    #include <windows.h>
    
    int global = 0;
    static int staticVal = 0;
    
    class Test
    { 
    public:
    	int m_val = 0;
    };
    
    int test1()
    { 
    	int tmp = 0;
    	for (int i = 0; i < 1000000000; i++)
    	{ 
    		tmp += i;
    	}
    	return tmp;
    }
    
    void test2(Test *t)
    { 
    	for (int i = 0; i < 1000000000; i++)
    	{ 
    		t->m_val += i;
    	}
    	return;
    }
    
    int test3()
    { 
    	for (int i = 0; i < 1000000000; i++)
    	{ 
    		global += i;
    	}
    	return global;
    }
    
    int test4()
    { 
    	for (int i = 0; i < 1000000000; i++)
    	{ 
    		staticVal += i;
    	}
    	return staticVal;
    }
    
    int main()
    { 
    	Test test;
    	unsigned long now, last;
    	now = last = GetTickCount();
    	test.m_val = test1();
    	now = GetTickCount();
    	std::cout << "test 1 val = " << test.m_val << ", now = " << now 
    		<< ", last = " << last << ", time cost = " << now - last << std::endl;
    
    	test.m_val = 0;
    	now = last = GetTickCount();
    	test2(&test);
    	now = GetTickCount();
    	std::cout << "test 2 val = " << test.m_val << ", now = " << now
    		<< ", last = " << last << ", time cost = " << now - last << std::endl;
    
    	test.m_val = 0;
    	now = last = GetTickCount();
    	test.m_val = test3();
    	now = GetTickCount();
    	std::cout << "test 3 val = " << test.m_val << ", now = " << now
    		<< ", last = " << last << ", time cost = " << now - last << std::endl;
    
    	test.m_val = 0;
    	now = last = GetTickCount();
    	test.m_val = test4();
    	now = GetTickCount();
    	std::cout << "test 4 val = " << test.m_val << ", now = " << now
    		<< ", last = " << last << ", time cost = " << now - last << std::endl;
    
    	getchar();
    	return 0;
    }

最终测试结果是test3()和test4()耗时相同且最短，test1()其次，test2()最慢。原因是test3()和test4()使用变量在编译时即开辟空间并存储在了.data字段中可以直接从内存读取到寄存器，而局部变量需要在栈中临时创建，存储于寄存器中，最慢的test2是因为每次均需要调用指针读取内存的值并存储。

### 3. 总结 ###

在写代码时，使用临时变量或者全局变量存储函数中需要计算的值，尽量减少多次指针访问和读写可以有效提高代码的执行速度