设置-O3时会发生什么

Question

I have the following code :

int main()
{
int i=0;
for(i=0;i<10000000;i++)
        //do something
return 0;
}

When I run this code using 'gcc -o file file.c' command, it is getting executed in 244 milli seconds and when I execute the same code using 'gcc -O3 file.c -o file' it is getting executed in 0 milli seconds(very little time which is shown as zero). I could not understand what optimization is done by -O3 flag in such a way that we get lot of speed in execution.

Any link,doc or an explaination regarding '-O3' optimization(apart from gcc.gnu.org, which I have seen and found the info given there is not sufficient to provide me a decent explaination and insight of what goes on) is much appreciated.

Answer 1

Ultimately we need to know what //do something does. You should paste your exact code that you used because technically the return statement would be executed as the body of the loop.

If you had something like this:

int main()
{
    int i=0;
    int junkInteger = 0;
    for(i=0; i < 5 ;i++)
    {
        junkInteger++;  // just to have a body
    }
    return 0;
}

With optimization, your loop may become unrolled like such and would thus cause your code size to become larger.

int main()
{
    int i=0;
    int junkInteger = 0;

    // loop gets unrolled
    junkInteger++;
    junkInteger++;
    junkInteger++;
    junkInteger++;
    junkInteger++;
    return 0;
}

When the loop gets unrolled, the low level assembly/machine code does not have to perform a Load, Store, Compare, and Branch to manage the loop.

You should try your code again with this:

int main()
{
    int i=0;
    for(i=0;i<10000000;i++)
    {
        //do something
    }
    return 0;
}

Compile it with both optimization on and off and examine the file sizes and even put in the compile option to generate assembly code. Usually with optimizations enabled, code size can become larger due to loop unrolls. Although the code could be optimized out since the body does nothing. Perhaps use a volatile loop counter to ensure it is not optimized out.