[Bug tree-optimization/101938] New: [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[Bug tree-optimization/101938] New: [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101938

            Bug ID: 101938
           Summary: [12 Regression] Wrong code with -fwrapv since
                    r12-2591-g2e96b5f14e402569
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: marxin at gcc dot gnu.org
                CC: aldyh at gcc dot gnu.org, amacleod at redhat dot com
  Target Milestone: ---

It's a test-case reduced from postgresql package:

$ cat postgresql.c 
#include <stdio.h>

typedef long long int int64;

#define INT64CONST(x)  (x##LL)
#define PG_INT64_MIN    (-INT64CONST(0x7FFFFFFFFFFFFFFF) - 1)

static int64
__attribute__((noipa))
int8gcd_internal(int64 arg1, int64 arg2)
{
        int64           swap;
        int64           a1,
                                a2;

        /*
         * Put the greater absolute value in arg1.
         *
         * This would happen automatically in the loop below, but avoids an
         * expensive modulo operation, and simplifies the special-case handling
         * for INT64_MIN below.
         *
         * We do this in negative space in order to handle INT64_MIN.
         */
        a1 = (arg1 < 0) ? arg1 : -arg1;
        a2 = (arg2 < 0) ? arg2 : -arg2;
        if (a1 > a2)
        {
                swap = arg1;
                arg1 = arg2;
                arg2 = swap;
        }

        /* Special care needs to be taken with INT64_MIN.  See comments above.
*/
        if (arg1 == PG_INT64_MIN)
        {
                if (arg2 == 0 || arg2 == PG_INT64_MIN)
    {
      __builtin_printf ("ret=out of range\n");
    }

                /*
                 * Some machines throw a floating-point exception for INT64_MIN
% -1,
                 * which is a bit silly since the correct answer is perfectly
                 * well-defined, namely zero.  Guard against this and just
return the
                 * result, gcd(INT64_MIN, -1) = 1.
                 */
                if (arg2 == -1) {
      __builtin_printf ("ret=%d\n", 1);
                        return 1;
    }
        }

        /* Use the Euclidean algorithm to find the GCD */
        while (arg2 != 0)
        {
                swap = arg2;
                arg2 = arg1 % arg2;
                arg1 = swap;
        }

        /*
         * Make sure the result is positive. (We know we don't have INT64_MIN
         * anymore).
         */
        if (arg1 < 0)
                arg1 = -arg1;

  __builtin_printf ("ret=%lld\n", arg1);
        return arg1;
}

int main()
{
  int8gcd_internal (-1, -9223372036854775808ULL);
  return 0;
}

$ gcc postgresql.c -fwrapv && ./a.out 
ret=1
$ gcc postgresql.c -O2 -fwrapv && ./a.out 
Floating point exception (core dumped)

[Bug tree-optimization/101938] [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101938

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |12.0
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2021-08-16
     Ever confirmed|0                           |1

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
As noted, started with r12-2591-g2e96b5f14e402569.

[Bug tree-optimization/101938] [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101938

--- Comment #2 from Martin Liška <marxin at gcc dot gnu.org> ---
Simplified test-case:

$ cat postgresql.c 
typedef long long int int64;
#define INT64CONST(x) (x##LL)
/* -9223372036854775808ULL */
#define INT64_MIN (-INT64CONST(0x7FFFFFFFFFFFFFFF) - 1)

static void __attribute__((noipa)) foo(int64 arg1, int64 arg2) {
  int64 a1 = -arg1;
  int64 a2 = (arg2 < 0) ? arg2 : -arg2;

  if (a1 > a2) {
    int64 swap = arg1;
    arg1 = arg2;
    arg2 = swap;
  }

  if (arg1 == INT64_MIN && arg2 == -1) return;

  __builtin_abort();
}

int main() {
  foo(-1, INT64_MIN);
  return 0;
}

[Bug tree-optimization/101938] [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101938

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1

[Bug tree-optimization/101938] [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101938

Aldy Hernandez <aldyh at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |aldyh at gcc dot gnu.org

--- Comment #3 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
Created attachment 51312
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51312&action=edit
proposed patch

[Bug tree-optimization/101938] [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101938

--- Comment #4 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
This is actually an oversight in the range-ops code.  In flag_wrapv
-TYPE_MIN_VALUE = TYPE_MIN_VALUE which is special cased in the ABS folding
routine, but not in operator_abs::op1_range().

Thank you for reporting and distilling this Martin.

[Bug tree-optimization/101938] [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101938

--- Comment #5 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Aldy Hernandez <aldyh@gcc.gnu.org>:

https://gcc.gnu.org/g:891bdbf2b0432b4aa3d3e76923617fcb4fd33cf6

commit r12-2944-g891bdbf2b0432b4aa3d3e76923617fcb4fd33cf6
Author: Andrew MacLeod <amacleod@redhat.com>
Date:   Tue Aug 17 10:50:56 2021 +0200

    Special case -TYPE_MIN_VALUE for flag_wrapv in operator_abs::op1_range.

    With flag_wrapv, -TYPE_MIN_VALUE = TYPE_MIN_VALUE which is
    unrepresentable.  We currently special case this in the ABS folding
    routine, but are missing similar treatment in operator_abs::op1_range.

    Tested on x86-64 Linux.

            PR tree-optimization/101938

    gcc/ChangeLog:

            * range-op.cc (operator_abs::op1_range): Special case
            -TYPE_MIN_VALUE for flag_wrapv.

    gcc/testsuite/ChangeLog:

            * gcc.dg/pr101938.c: New test.

[Bug tree-optimization/101938] [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101938

Aldy Hernandez <aldyh at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #6 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
fixed in trunk

[Bug tree-optimization/101938] [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101938

--- Comment #7 from Martin Liška <marxin at gcc dot gnu.org> ---
> Thank you for reporting and distilling this Martin.

You're welcome, it was pretty fun isolating that!
Thanks for the hot fix.

[Bug tree-optimization/101938] [12 Regression] Wrong code with -fwrapv since r12-2591-g2e96b5f14e402569

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101938

--- Comment #8 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
(In reply to Martin Liška from comment #7)
> > Thank you for reporting and distilling this Martin.
> 
> You're welcome, it was pretty fun isolating that!
> Thanks for the hot fix.

That was all Andrew!  I just wrote comments, tests, and took the credit ;-).