[Bug tree-optimization/111583] New: [13/14 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[Bug tree-optimization/111583] New: [13/14 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[shaohua.li at inf dot ethz.ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111583

            Bug ID: 111583
           Summary: [13/14 Regression] Wrong code at -Os on
                    x86_64-linux-gnu since r13-3281-g6cc3394507
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: shaohua.li at inf dot ethz.ch
                CC: amacleod at redhat dot com
  Target Milestone: ---

gcc at -Os produced the wrong code.

Bisected to r13-3281-g6cc3394507

Compiler explorer: https://godbolt.org/z/8GM9YvMKb

$ cat a.c
int printf(const char *, ...);
int b, c, d;
char e;
short f;
const unsigned short **g;
char h(char k) {
  if (k)
    return '0';
  return 0;
}
int l() {
  b = 0;
  return 1;
}
static short m(unsigned k) {
  const unsigned short *n[65];
  g = &n[4];
  k || l();
  long a = k;
  char i = 0;
  unsigned long j = k;
  while (j--)
    *(char *)a++ = i;
  c = h(d);
  f = k;
  return 0;
}
int main() {
  long o = (e < 0) << 5;
  m(o);
  printf("%d\n", f);
}
$
$ gcc -O0 -fsanitize=address,undefined a.c && ./a.out
0
$ gcc -Os a.c && ./a.out
32
$

[Bug tree-optimization/111583] [13/14 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111583

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |13.3

[Bug tree-optimization/111583] [13/14 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111583

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2023-09-25

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed.

The problem is a latent bug in ldist.
It turns:
```
  <bb 5> [local count: 955630224]:
  a_5 = a_4 + 1;
  a.4_6 = (char *) a_4;
  *a.4_6 = 0;

  <bb 6> [local count: 1073741824]:
  # a_4 = PHI <a_2(4), a_5(5)>
  # j_7 = PHI <j_3(4), j_8(5)>
  j_8 = j_7 + 18446744073709551615;
  if (j_7 != 0)
    goto <bb 5>; [89.00%]
  else
    goto <bb 7>; [11.00%]
```

Into:
```
  a_2 = (long int) k_1(D);
  j_3 = (long unsigned int) k_1(D);
  _23 = (sizetype) k_1(D);
  _25 = (char *) a_2;
  __builtin_memset (_25, 0, _23);
```

Which then basically says k!=0 as _25 can't be a null pointer.

[Bug tree-optimization/111583] [13/14 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111583

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 111631 has been marked as a duplicate of this bug. ***

[Bug tree-optimization/111583] [13/14 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111583

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
A conservative fix would be to make sure the memory accesses for the
memset/memcpy recognition are before the loop exit.  Note since -Os disables
loop header copying in most cases this will remove quite a bit of memset/memcpy
loop detection there.  The alternative would be to guard the calls.

[Bug tree-optimization/111583] [13/14 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111583

--- Comment #4 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:962ca7149d652e4077a2259886e5cd2ea3cea0ab

commit r14-4329-g962ca7149d652e4077a2259886e5cd2ea3cea0ab
Author: Richard Biener <rguenther@suse.de>
Date:   Fri Sep 29 11:08:18 2023 +0200

    tree-optimization/111583 - loop distribution issue

    The following conservatively fixes loop distribution to only
    recognize memset/memcpy and friends when at least one element
    is going to be processed.  This avoids having an unconditional
    builtin call in the IL that might imply the source and destination
    pointers are non-NULL when originally pointers were not always
    dereferenced.

    With -Os loop header copying is less likely to ensure this.

            PR tree-optimization/111583
            * tree-loop-distribution.cc (find_single_drs): Ensure the
            load/store are always executed.

            * gcc.dg/tree-ssa/pr111583-1.c: New testcase.
            * gcc.dg/tree-ssa/pr111583-2.c: Likewise.

[Bug tree-optimization/111583] [13 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111583

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[13/14 Regression] Wrong    |[13 Regression] Wrong code
                   |code at -Os on              |at -Os on x86_64-linux-gnu
                   |x86_64-linux-gnu since      |since r13-3281-g6cc3394507
                   |r13-3281-g6cc3394507        |
      Known to fail|                            |13.2.0
      Known to work|                            |14.0

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed on trunk.  The issue is latent for longer but at least pre-ranger we
probably were not good enough to extract range info from this.

[Bug tree-optimization/111583] [13 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111583

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug tree-optimization/111583] [13 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111583

--- Comment #6 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-13 branch has been updated by Richard Biener
<rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:b28be88c311f1bc4c62309906177ab3d97800deb

commit r13-8041-gb28be88c311f1bc4c62309906177ab3d97800deb
Author: Richard Biener <rguenther@suse.de>
Date:   Fri Sep 29 11:08:18 2023 +0200

    tree-optimization/111583 - loop distribution issue

    The following conservatively fixes loop distribution to only
    recognize memset/memcpy and friends when at least one element
    is going to be processed.  This avoids having an unconditional
    builtin call in the IL that might imply the source and destination
    pointers are non-NULL when originally pointers were not always
    dereferenced.

    With -Os loop header copying is less likely to ensure this.

            PR tree-optimization/111583
            * tree-loop-distribution.cc (find_single_drs): Ensure the
            load/store are always executed.

            * gcc.dg/tree-ssa/pr111583-1.c: New testcase.
            * gcc.dg/tree-ssa/pr111583-2.c: Likewise.

    (cherry picked from commit 962ca7149d652e4077a2259886e5cd2ea3cea0ab)

[Bug tree-optimization/111583] [13 Regression] Wrong code at -Os on x86_64-linux-gnu since r13-3281-g6cc3394507

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111583

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
      Known to work|                            |13.2.1
             Status|ASSIGNED                    |RESOLVED

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.