[Bug c/110491] New: Wrong code at -O2 on x86_64-linux-gnu (a recent regression)

[Bug c/110491] New: Wrong code at -O2 on x86_64-linux-gnu (a recent regression)

[shaohua.li at inf dot ethz.ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110491

            Bug ID: 110491
           Summary: Wrong code at -O2 on x86_64-linux-gnu (a recent
                    regression)
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: shaohua.li at inf dot ethz.ch
  Target Milestone: ---

gcc at -O2 produced wrong code.

Compiler explorer: https://godbolt.org/z/cbM1q13G7

$ cat a.c
int printf(const char *, ...);
int a, c, d, e;
short b;
void f(int *g) { c &= *g; }
void h(void);
void i() {
  a = 1;
  h();
  f(&a);
}
void h() {
  int *j = &c;
  *j = 5;
k:
  for (; 4 + b <= 0;)
    ;
  for (; d;) {
    c = e == 0;
    goto k;
  }
}
int main() {
  i();
  printf("%d\n", c);
}
$
$ gcc-tk -O0 a.c &&./a.out
1
$ gcc-tk -O2 a.c && ./a.out
-2003318871
$
$ gcc-tk -v
Using built-in specs.
COLLECT_GCC=gcc-tk
COLLECT_LTO_WRAPPER=/zdata/shaoli/compilers/ccbuilder-compilers/gcc-5f590ee3174cf6058ac882c3a84a96ae639349c8/libexec/gcc/x86_64-pc-linux-gnu/14.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../configure --disable-multilib --disable-bootstrap
--enable-languages=c,c++
--prefix=/zdata/shaoli/compilers/ccbuilder-compilers/gcc-5f590ee3174cf6058ac882c3a84a96ae639349c8
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 14.0.0 20230629 (experimental) (GCC)
$

[Bug c/110491] Wrong code at -O2 on x86_64-linux-gnu (a recent regression)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110491

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |DUPLICATE

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Dup of bug 110228.



  # RANGE [irange] int [0, 1] NONZERO 0x1
  # c_lsm.10_3 = PHI <_17(D)(2), _11(5)>
...
  if (c_lsm_flag.11_16 != 0)
    goto <bb 8>; [66.67%]
  else
    goto <bb 7>; [33.33%]
;;    succ:       7 [33.3% (adjusted)]  count:357913944 (estimated locally)
(FALSE_VALUE,EXECUTABLE)
;;                8 [66.7% (adjusted)]  count:715827880 (estimated locally)
(TRUE_VALUE,EXECUTABLE)

;;   basic block 7, loop depth 0, count 357913944 (estimated locally), maybe
hot
;;    prev block 6, next block 8, flags: (NEW)
;;    pred:       6 [33.3% (adjusted)]  count:357913944 (estimated locally)
(FALSE_VALUE,EXECUTABLE)
;;    succ:       8 [always]  count:357913944 (estimated locally) (FALLTHRU)

;;   basic block 8, loop depth 0, count 1073741824 (estimated locally), maybe
hot
;;    prev block 7, next block 1, flags: (NEW, REACHABLE, VISITED)
;;    pred:       7 [always]  count:357913944 (estimated locally) (FALLTHRU)
;;                6 [66.7% (adjusted)]  count:715827880 (estimated locally)
(TRUE_VALUE,EXECUTABLE)
  # RANGE [irange] int [0, 1] NONZERO 0x1
  # prephitmp_27 = PHI <1(7), c_lsm.10_3(6)>

*** This bug has been marked as a duplicate of bug 110228 ***

[Bug c/110491] Wrong code at -O2 on x86_64-linux-gnu (a recent regression)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110491

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
             Status|RESOLVED                    |ASSIGNED
         Resolution|DUPLICATE                   |---
   Last reconfirmed|                            |2023-07-04

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
So the issue here is phiopt4 doing

   <bb 6> [local count: 1073741824]:
-  if (c_lsm_flag.14_16 != 0)
-    goto <bb 8>; [66.67%]
-  else
-    goto <bb 7>; [33.33%]
-
-  <bb 7> [local count: 357913944]:
-
-  <bb 8> [local count: 1073741824]:
+  _26 = ~c_lsm_flag.14_16;
+  _14 = (int) _26;
   # RANGE [irange] int [0, 1] NONZERO 0x1
-  # prephitmp_27 = PHI <1(7), c_lsm.13_3(6)>
-  c = prephitmp_27;
+  _4 = c_lsm.13_3 | _14;
+  c = _4;
   return;

via

phiopt match-simplify trying:
       c_lsm_flag.14_16 != 0 ? c_lsm.13_3 : 1
Matching expression match.pd:1991, gimple-match-5.cc:23
Matching expression match.pd:1991, gimple-match-5.cc:23
Matching expression match.pd:1991, gimple-match-5.cc:23
Matching expression match.pd:1948, gimple-match-7.cc:20
Matching expression match.pd:2480, gimple-match-4.cc:35
Matching expression match.pd:2483, gimple-match-3.cc:66
Matching expression match.pd:2490, gimple-match-2.cc:58
Applying pattern match.pd:6459, gimple-match-10.cc:12767
Applying pattern match.pd:1378, gimple-match-5.cc:7352
Applying pattern match.pd:1886, gimple-match-1.cc:1232
Applying pattern match.pd:4748, gimple-match-4.cc:15651
Folded into the sequence:
_13 = (int) c_lsm_flag.14_16;
_26 = ~c_lsm_flag.14_16;
_14 = (int) _26;
_4 = c_lsm.13_3 | _14;
Removing basic block 7

but

  <bb 3> [local count: 9761289309]:
  # RANGE [irange] int [0, 1] NONZERO 0x1
  # c_lsm.13_3 = PHI <_17(D)(2), _11(5)>

so we have an uninitialized use here.  The following fixes this.

diff --git a/gcc/tree-ssa-phiopt.cc b/gcc/tree-ssa-phiopt.cc
index 31a7c39e405..467c9fd108a 100644
--- a/gcc/tree-ssa-phiopt.cc
+++ b/gcc/tree-ssa-phiopt.cc
@@ -785,6 +785,13 @@ match_simplify_replacement (basic_block cond_bb,
basic_block middle_bb,
       arg_false = arg0;
     }

+  /* Do not make conditional undefs unconditional.  */
+  if ((TREE_CODE (arg_true) == SSA_NAME
+       && ssa_name_maybe_undef_p (arg_true))
+      || (TREE_CODE (arg_false) == SSA_NAME
+         && ssa_name_maybe_undef_p (arg_false)))
+    return false;
+
   tree type = TREE_TYPE (gimple_phi_result (phi));
   result = gimple_simplify_phiopt (early_p, type, stmt,
                                   arg_true, arg_false,

[Bug c/110491] Wrong code at -O2 on x86_64-linux-gnu (a recent regression)

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110491

--- Comment #3 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:819285ef10a87d663f8c181c06aa88d1d9f75aed

commit r14-2296-g819285ef10a87d663f8c181c06aa88d1d9f75aed
Author: Richard Biener <rguenther@suse.de>
Date:   Tue Jul 4 12:52:27 2023 +0200

    tree-optimization/110491 - PHI-OPT and undefs

    The following makes sure to not make conditional undefs in PHI arguments
    unconditional by folding cond ? arg1 : arg2.

            PR tree-optimization/110491
            * tree-ssa-phiopt.cc (match_simplify_replacement): Check
            whether the PHI args are possibly undefined before folding
            the COND_EXPR.

            * gcc.dg/torture/pr110491.c: New testcase.

[Bug c/110491] Wrong code at -O2 on x86_64-linux-gnu (a recent regression)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110491

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

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

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.