低级C/C++性能?

Luk*_*uka 5 c c++ micro-optimization

更新:我只是设法击败我自己的32代码:

void test(char *file_char, unsigned int size)
{
    char* file_ = file_char;
    char* size_x = file_char+size;
    char to_find = 0;

    for(unsigned int i = 0; i < 10000; i++)
    {
        file_char = file_;

        while(*file_char++ != to_find);//skip all characters till we find a 0

        if(*file_char)//some if in order to avoid compiler removing our test code
            cout << "found";
    }
}
Run Code Online (Sandbox Code Playgroud)

上面的代码要求0在数组中至少出现一次,否则会出现错误但是它比if代码快一点,而且更紧凑.

有没有办法让上面的代码更快?(有一个char数组并试图找到一个char出现的位置)?

我写了一些代码,我真的很困惑.

在里面:

int main()
{
    FILE *file;
    file = fopen("C:\\data.txt", "rb");

    static const int size = 60000;

    char *file_char = (char*)malloc(size);

    unsigned int i = 0;
    while(i < size)
        fread(&file_char[i++], 1, 1, file);

    clock_t clock_ = clock();
    test(file_char, size);
    std::cout << ((double)clock()-clock_)/1000;
    return 0;
}
Run Code Online (Sandbox Code Playgroud)

下面的代码需要3.5秒才能执行:

void test(char *file_char, unsigned int size)
{
    for(unsigned int i = 0; i < 100000; i++)
    {
        unsigned int pos = 0;
        char to_find = 0;
        while(pos < size)
            if(file_char[pos++] == to_find)
                std::cout << "found";
    }
}
Run Code Online (Sandbox Code Playgroud)

但是下面的代码需要1.8秒,HALF时间!

void test(char *file_char, unsigned int size)
{
    for(unsigned int i = 0; i < 100000; i++)
    {
        unsigned int pos = 0;
        char to_find = 0;
        while(pos < size)
        {
            if(file_char[pos] == to_find)
                std::cout << "found";
            else if(file_char[pos+1] == to_find)
                std::cout << "found";
            else if(file_char[pos+2] == to_find)
                std::cout << "found";
            else if(file_char[pos+3] == to_find)
                std::cout << "found";
            else if(file_char[pos+4] == to_find)
                std::cout << "found";
            else if(file_char[pos+5] == to_find)
                std::cout << "found";
            else if(file_char[pos+6] == to_find)
                std::cout << "found";
            else if(file_char[pos+7] == to_find)
                std::cout << "found";
            else if(file_char[pos+8] == to_find)
                std::cout << "found";
            else if(file_char[pos+9] == to_find)
                std::cout << "found";
            else if(file_char[pos+10] == to_find)
                std::cout << "found";
            else if(file_char[pos+11] == to_find)
                std::cout << "found";
            else if(file_char[pos+12] == to_find)
                std::cout << "found";
            else if(file_char[pos+13] == to_find)
                std::cout << "found";
            else if(file_char[pos+14] == to_find)
                std::cout << "found";
            else if(file_char[pos+15] == to_find)
                std::cout << "found";
            else if(file_char[pos+16] == to_find)
                std::cout << "found";
            else if(file_char[pos+17] == to_find)
                std::cout << "found";
            else if(file_char[pos+18] == to_find)
                std::cout << "found";
            else if(file_char[pos+19] == to_find)
                std::cout << "found";
            else if(file_char[pos+20] == to_find)
                std::cout << "found";
            else if(file_char[pos+21] == to_find)
                std::cout << "found";
            else if(file_char[pos+22] == to_find)
                std::cout << "found";
            else if(file_char[pos+23] == to_find)
                std::cout << "found";
            else if(file_char[pos+24] == to_find)
                std::cout << "found";
            else if(file_char[pos+25] == to_find)
                std::cout << "found";
            else if(file_char[pos+26] == to_find)
                std::cout << "found";
            else if(file_char[pos+27] == to_find)
                std::cout << "found";
            else if(file_char[pos+28] == to_find)
                std::cout << "found";
            else if(file_char[pos+29] == to_find)
                std::cout << "found";
            else if(file_char[pos+30] == to_find)
                std::cout << "found";
            else if(file_char[pos+31] == to_find)
                std::cout << "found";

            pos+=32;
        }
    }
}
Run Code Online (Sandbox Code Playgroud)

我使用的是Visual Studio 2012 x64,程序从不会cout任何东西,因为没有char是0.如何解释?如何在不使用32 ifs的情况下归档相同的性能?

编辑1:如果我创建了64个ifs,那么32 ifs版本的速度没有增加.

编辑2:如果我删除else并离开ifs程序需要4秒.

现在,如何解释上述不合理的结果?

MrZ*_*bra 9

你的循环基本上由两个比较组成:pos < sizefile_char[pos] == to_find.通过展开循环,您可以将比较次数从2*大小减少到(大小+大小/ 32).


Iri*_*ces 5

我认为这两个代码是不同的.

在第一个中,您每次都会检查"if"比较.

在第二个,如果第一个是好的,你跳过以下所有!(因为其他)所以你节省了很多比较(但缺少支票).

要获得相同的代码,您必须删除所有"其他".


man*_*lio 4

我做了一些测试来确定。

使用 g++(在 Linux 和 Windows 下),我得到与 Visual Studio 相同的结果:

版本 1(无显式循环展开)

g++ -O37.5秒

版本 2(显式循环展开)

g++ -O32.1秒

但打开-funroll-loops选项后(通常默认情况下不启用此优化,因为它可能会也可能不会使其运行得更快):

版本 1(无显式循环展开)

g++ -O3 -funroll-loops2.2秒

版本 2(显式循环展开)

g++ -O3 -funroll-loops2.2秒

所以这与循环展开有关。

编辑

您可以更改最后一个示例以显式插入哨兵,例如:

int main()
{
  static const int size = 60000;

  char *file_char = (char*)malloc(size+1);  // The last element is the sentry

  // ...Fill file_char[]...

  file_char[size] = 0;  // the sentry

  // ...
}
Run Code Online (Sandbox Code Playgroud)

所以test函数不会失败(当然你必须检查是否找到了哨兵或“好”零,但这只是一个如果)。

版本 3(哨兵)

g++ -O30.68秒

g++ -O3 -funroll-loops0.72秒