Thread: -funroll-all-loops

I was wondering how exactly GCC unrolls loops. Like I can imagine if you have:

Code:

for (int i=0; i<100; ++i) {a();};

it can just put a 100nd times a(). But what if you have:

Code:

int max;
....
for (int i=0; i<max; ++i) {a();};

so the compiler won't know the value of max. Can this loop be unrolled with -O3 -funroll-all-loops? Is there a way for the compiler to optimize this loop in terms of speed?
If you have two loops like:

Code:

for (int j=0; j<maxj; ++j) 
    for (int i=0; i<maxi; ++i) {
        a();
    };

is there any way to merge somehow the loops to gain speed?

I did.... after -O3 -funroll-all-loops is really to understand anything :P Well, my code is large, so I ll make a code only with the loop and see. Thanx for the advise, will get back if I don't figure it out

It doesn't seem to unroll at all the loops. I do this:

Code:

#include <stdio.h>
#include <stdlib.h>

int main()
{
    int k, j=0;
    for (k=0; k < 100; ++k)
	    j++;
}

gcc -S -c test.c
with -funroll-all-loops it generates the same code.

With -O3 I just make no sense, though there are no jump commands at all