replace_copy的效率问题

[szn_409]

{
//replace:前两个参数为目标序列，第三个参数为被替换值，第四参数为替换值
const int size = 1e7;
vector<int> vecInt;
vecInt.resize(size);

fill(vecInt.begin(), vecInt.end(), 10);

int time0 = GetTickCount();
replace(vecInt.begin(), vecInt.end(), 10, 20);
time0 = GetTickCount() - time0; //time0 = 16

time0 = GetTickCount();
for (int i = 0; i < size; ++i)
{
vecInt[i] == 10 ? vecInt[i] = 20 : 0;
}
time0 = GetTickCount() - time0; //time0 = 468

//replace_copy的速度令人发指,比手工操作还慢
vector<int> vecInt1;
time0 = GetTickCount();
replace_copy(vecInt.begin(), vecInt.end(), back_inserter(vecInt1), 20, 30);
time0 = GetTickCount() - time0; //time0 = 3058

vector<int> vecInt2;
time0 = GetTickCount();
for (int i = 0; i < size; ++i)
{
if (vecInt[i] == 20)
{
vecInt2.push_back(30);
}
else
{
vecInt2.push_back(vecInt[i]);
}
}
time0 = GetTickCount() - time0; //time0 = 2808
printf("\n");
}

如上代码，为什么replace_copy的效率还不如手写的效率高？

[dustpg]

没有仔细看lz的代码, 工具lz的代码, Release模式第三个比第四个快1/3(100毫秒与150毫米)， 而lz的是4秒, 所以怀疑lz没有开优化，如果不开优化, STL一般来说很慢