An interesting bug on GCC 9 and above

Code:

---

#include <stdio.h>

int main ( void )
{
  unsigned int a;
  unsigned int b;
  unsigned int c;

  c = 0;
  for ( a = 0; a < 10; a++ )
    for ( b = 0; b < 2; b++ )
    {
      c++;
         
      printf( "a=%u, c=%u\n", a, c );

      if ( c < a )
        return 123;
    }

  return 0;
}

---

On GCC 9+ this code will print on 3 lines and return 123 if you compile with any optimization options (any of -O options, except, of course -O0). As you can see there is no overflow and c will never be less than a. It should print 20 lines and return 0.

Change the comparison to b < N, where N != 2, in the inner loop and will work correctly. Or change the variables to 'int' (signed) and will work correctly.

This bug isn't present on GCC 8 and below.

Quote:

---

Originally Posted by flp1969

Code:

---

#include <stdio.h>

int main ( void )
{
  unsigned int a;
  unsigned int b;
  unsigned int c;

  c = 0;
  for ( a = 0; a < 10; a++ )
    for ( b = 0; b < 2; b++ )
    {
      c++;
         
      printf( "a=%u, c=%u\n", a, c );

      if ( c < a )
        return 123;
    }

  return 0;
}

---

On GCC 9+ this code will print on 3 lines and return 123 if you compile with any optimization options (any of -O options, except, of course -O0). As you can see there is no overflow and c will never be less than a. It should print 20 lines and return 0.

Change the comparison to b < N, where N != 2, in the inner loop and will work correctly. Or change the variables to 'int' (signed) and will work correctly.

This bug isn't present on GCC 8 and below.

---

I am running gcc version 10.2.1, on Debian testing, and -O1, -O2, and -O3, as well as with no -O options, all with -std=c18, and the code runs as you expect it should, and all return 0.

I cannot comment on this any further.

Well that's pretty weird.
Compiler Explorer

Lot's of compilers to choose from.

I'm on Linux Mint 19.3 with GCC 9.3.0, and I can confirm it happens for me too. I'm trying to find any undefined behavior in the code but can't so far. Definitely looks like a bug in GCC 9.

Correction it did have the bug; I forgot to set optimize option on first test

Code:

---

gcc.exe (Rev5, Built by MSYS2 project) 10.3.0

---

Tim S.

Very strange. You can even start c at 1000. Interesting how changing unsigned to int fixes it.

Code:

---

#include <stdio.h>
 
int main()
{
    unsigned a, b, c = 1000;

    for (a = 0; a < 5; ++a)
    {
        for (b = 0; b < 2; ++b)
        {
            ++c;
            printf("a=%u, c=%u\n", a, c);
            if (c < a)
            {
                printf("Premature termination\n");
                return 0;
            }
        }
    }
 
    printf("Normal termination\n");
    return 0;
}

---

I did not see the problem with "-Og" can anyone else confirm this?
MSys2 GCC 10.3.0

-Og enables optimizations that do not interfere with debugging.
It would not include unrolling loops, which is at least part of what's happening here, albeit incorrectly.

Quote:

---

Originally Posted by john.c

-Og enables optimizations that do not interfere with debugging.
It would not include unrolling loops, which is at least part of what's happening here, albeit incorrectly.

---

That is my guess; but, I have found my guess are often wrong about which optimize option causes a GCC program bug/crash.

Tim S.

I looked at the generated assembly (with -S) and it has unrolled the loop, but only for 3 iterations. However, just using -funroll-loops doesn't cause the problem.