使用-O3的osx上的g ++优化错误

Question

The following code produces the wrong exit code when compiled with -O3. I think the inner loop is being optimized away incorrectly. With -O2 or -fno-inline it works. Producing a simpler example is difficult because any small changes and the bug disappears.

Compiled with:

/usr/bin/g++ -O3 -o bugexample bugexample.cpp

Code:

#include <vector>

int test(std::vector<char>& a, int& b)
{
    std::vector<int> z;
    z.push_back(10);
    z.push_back(10);

    int d = (int)a.size();

    int x = 1;
    for (int j = 0; j < 2; j++)
    {
        int c = j - 1;

        for (int i = 0; i < d; i++) 
        {
            if (j == 0)
            {
            }
            else if (i == 0)
            {
            }
            else
            {
                if (a[j] == a[i - 1])
                {
                    b = c + 1;
                    x = 2;
                }
                z[i] = 1;
            }
        }
    }

    return x;
}


int main(int argc, char* argv[])
{
    std::vector<char> a;
    a.push_back('a');
    a.push_back('a');
    int b = 1;
    return test(a,b);
}

Compiler version:

/usr/bin/g++ -v
Using built-in specs.
Target: i686-apple-darwin10
Configured with: /var/tmp/gcc/gcc-5666.3~123/src/configure --disable-checking --enable-werror --prefix=/usr --mandir=/share/man --enable-languages=c,objc,c++,obj-c++ --program-transform-name=/^[cg][^.-]*$/s/$/-4.2/ --with-slibdir=/usr/lib --build=i686-apple-darwin10 --program-prefix=i686-apple-darwin10- --host=x86_64-apple-darwin10 --target=i686-apple-darwin10 --with-gxx-include-dir=/include/c++/4.2.1
Thread model: posix
gcc version 4.2.1 (Apple Inc. build 5666) (dot 3)

Interested in any insite, or proof that its my fault.

Edit: The exit code produced is 1, whereas it should b 2.

Answer 1

Mmm is this some kind of obfuscated code contest?

As far as I can tell you're trying to do some kind of palindrome test on the input vector. Only,

• the loop var j has a hardcoded upperbound of 2 (which should probably have been a.size() as well?)
• you only return the check of the last position
• you had all kinds of redundant conditions
• you had gratuitous non-const arguments
• you had unused z vector
• you had unnecessary use of int for bool (1=>false - not found, 2=>true - found)
• you had unnecessary use of out parameter b ; I replaced that bool return type with the value of b (with b==-1 indicating no match found)

When simplifying the code for these things, I get this code, and (like your own code) it behaves identically for all optimization levels on g++ 4.6.1:

#include <vector>

int test(const std::vector<char>& a)
{
    /* int j = 1; // was: for (int j = 1; j < 2; j++) */

    for (int i = a.size()-1; i > 1; i--) 
        if (a[1] == a[i - 1])
            return 1;

    return -1;
}


int main(int argc, char* argv[])
{
    std::vector<char> a(2, 'a');
    int b = test(a);

    return b==-1? 1 : 2;
}