Vector遍历性能对比

偶然发现，遍历vector时，使用两种看上去没什么差别的方法，性能上却有很大的差异。

#include <iostream>
#include <vector>

class CPoint{
public:
    int x;
    int y;
};

int VectorRead_0(const CPoint* arr, size_t size)
{
    int result = 0;
    for(size_t i = 0; i < size; i ++){
        result += (arr[i].x + arr[i].y);
    }
    return result;
}
int VectorRead_1(const vector<CPoint> &arr)
{
    int result = 0;
    for(size_t i = 0; i < arr.size(); i ++){
        result += (arr[i].x + arr[i].y);
    }
    return result;
}
int VectorRead_2(const vector<CPoint> &arr)
{
    size_t size = arr.size();
    int result = 0;
    for(size_t i = 0; i < size; i ++){
        result += (arr[i].x + arr[i].y);
    }
    return result;
}
int VectorRead_3(const vector<CPoint> &arr)
{
    size_t size = arr.size();
    const CPoint* pArr = &arr[0];
    int result = 0;
    for(size_t i = 0; i < size; i ++){
        result += (pArr[i].x + pArr[i].y);
    }
    return result;
}

void TestVectorRead()
{
    const int arrSize = 100000;
    vector<CPoint> arr(arrSize);
    int        i, nLoopCount = 100000;
    DWORD    start,end;
    int    result;

    for(i = 0; i < arrSize; i ++){
        arr[i].x = i % 3;
        arr[i].y = i % 4;
    }

    start = ::GetTickCount();
    result = 0;
    for (i = 0; i < nLoopCount; i++){
        result += VectorRead_0(&arr[0], arrSize);
    }
    end = ::GetTickCount();
    cout << "VectorRead_0 takes: " << end - start << " result: " << result << std::endl;

    start = ::GetTickCount();
    result = 0;
    for (i = 0; i < nLoopCount; i++){
        result += VectorRead_1(arr);
    }
    end = ::GetTickCount();
    cout << "VectorRead_1 takes: " << end - start << " result: " << result << std::endl;

    start = ::GetTickCount();
    result = 0;
    for (i = 0; i < nLoopCount; i++){
        result += VectorRead_2(arr);
    }
    end = ::GetTickCount();
    cout << "VectorRead_2 takes: " << end - start << " result: " << result << std::endl;

    start = ::GetTickCount();
    result = 0;
    for (i = 0; i < nLoopCount; i++){
        result += VectorRead_3(arr);
    }
    end = ::GetTickCount();
    cout << "VectorRead_3 takes: " << end - start << " result: " << result << std::endl;
}

执行结果（运行环境：ThinkPad T430，VS2013）：

VectorRead_0 takes: 5663 result: -769903776
VectorRead_1 takes: 0 result: -769903776
VectorRead_2 takes: 0 result: -769903776
VectorRead_3 takes: 0 result: -769903776

可见，VectorRead_0性能非常糟糕，很可能编译器对vector有特殊的优化。

另外，用同样的方式测试了vector在写入情况下的性能，几种方法在性能上并没有明显差异。