[Bug tree-optimization/109546] New: [13 Regression] Missed Dead Code Elimination when using __builtin_unreachable

[Bug tree-optimization/109546] New: [13 Regression] Missed Dead Code Elimination when using __builtin_unreachable

[theodort at inf dot ethz.ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109546

            Bug ID: 109546
           Summary: [13 Regression]  Missed Dead Code Elimination when
                    using __builtin_unreachable
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: theodort at inf dot ethz.ch
  Target Milestone: ---

cat input.c 

void foo(void);
static int a, c;
static int *b = &a;
static int **d = &b;
void __assert_fail() __attribute__((__noreturn__));
int main() {
  int *e = *d;
  if (e == &a || e == &c);
  else {
  __assert_fail();
  }
  if (e == &a || e == &c);
  else
    foo();
}

both gcc 12 and trunk at -O3 generate the following code:

main:
        movq    b(%rip), %rax
        cmpq    $c, %rax
        je      .L2
        cmpq    $a, %rax
        jne     .L7
.L2:
        xorl    %eax, %eax
        ret
.L7:
        pushq   %rax
        xorl    %eax, %eax
        call    __assert_fail
b:
        .quad   a

the call to foo is eliminated but the call to __assert_fail is missed, even
though 
both if statement conditions are identical. 

If add a __builtin_unreachable before the call to __assert_fail the opposite
happens:

void foo(void);
static int a, c;
static int *b = &a;
static int **d = &b;
void assert_fail() __attribute__((__noreturn__));
int main() {
  int *e = *d;
  if (e == &a || e == &c);
  else {
    __builtin_unreachable(); 
  assert_fail();
  }
  if (e == &a || e == &c);
  else
    foo();
}

gcc-trunk -O3 generates:

main:
        movq    b(%rip), %rax
        cmpq    $c, %rax
        je      .L4
        cmpq    $a, %rax
        je      .L4
        pushq   %rax
        call    foo
        xorl    %eax, %eax
        popq    %rdx
        ret
.L4:
        xorl    %eax, %eax
        ret
b:
        .quad   a

the call to __assert_fail is now properly eliminated but now the call to foo is
missed. 

This is a regression as gcc-12 generates: 

main:
        xorl    %eax, %eax
        ret

[Bug tree-optimization/109546] [13/14 Regression] Missed Dead Code Elimination when using __builtin_unreachable

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109546

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[13 Regression]  Missed     |[13/14 Regression]  Missed
                   |Dead Code Elimination when  |Dead Code Elimination when
                   |using __builtin_unreachable |using __builtin_unreachable
           Keywords|                            |missed-optimization
   Target Milestone|---                         |13.0

[Bug tree-optimization/109546] [13/14 Regression] Missed Dead Code Elimination when using __builtin_unreachable since r13-3596-ge7310e24b1c0ca

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109546

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amacleod at redhat dot com,
                   |                            |marxin at gcc dot gnu.org
            Summary|[13/14 Regression]  Missed  |[13/14 Regression]  Missed
                   |Dead Code Elimination when  |Dead Code Elimination when
                   |using __builtin_unreachable |using __builtin_unreachable
                   |                            |since
                   |                            |r13-3596-ge7310e24b1c0ca

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
Started with r13-3596-ge7310e24b1c0ca.

[Bug tree-optimization/109546] [13/14 Regression] Missed Dead Code Elimination when using __builtin_unreachable since r13-3596-ge7310e24b1c0ca

[amacleod at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109546

--- Comment #2 from Andrew Macleod <amacleod at redhat dot com> ---
*** Bug 109588 has been marked as a duplicate of this bug. ***

[Bug tree-optimization/109546] [13/14 Regression] Missed Dead Code Elimination when using __builtin_unreachable since r13-3596-ge7310e24b1c0ca

[amacleod at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109546

--- Comment #3 from Andrew Macleod <amacleod at redhat dot com> ---
Patch in testing.

when deciding to  fold condition of builtin_unreachable, VRP is failing to
recognize that given 

  if (ptr == &x)
    __builtin_unreachable ()

&x is not a representable range, and thus should not be eliminated early even
though it satisfies the other criteria.

it just sees it as a symbolic value == CONSTANT  and thinks its smart enough
when it isn't.

[Bug tree-optimization/109546] [13/14 Regression] Missed Dead Code Elimination when using __builtin_unreachable since r13-3596-ge7310e24b1c0ca

[amacleod at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109546

--- Comment #4 from Andrew Macleod <amacleod at redhat dot com> ---
And I think the first part of this is a duplicate of another PR  from GCC12
time (I dont know which one) whereby we are not making staticly initialized
values available in main().  so VRP never sees that d == &b, and can do nothing
with the first condition, thus leaving the call to __assert_fail.

[Bug tree-optimization/109546] [13/14 Regression] Missed Dead Code Elimination when using __builtin_unreachable since r13-3596-ge7310e24b1c0ca

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109546

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

https://gcc.gnu.org/g:f828503eeb79ad1f1ada6db7deccc5abcc2f3ca3

commit r14-160-gf828503eeb79ad1f1ada6db7deccc5abcc2f3ca3
Author: Andrew MacLeod <amacleod@redhat.com>
Date:   Fri Apr 21 15:03:43 2023 -0400

    Do not fold ADDR_EXPR conditions leading to builtin_unreachable early.

    Ranges can not represent &var globally yet, so we cannot fold these
    expressions early or we lose the __builtin_unreachable information.

            PR tree-optimization/109546
            gcc/
            * tree-vrp.cc (remove_unreachable::remove_and_update_globals): Do
            not fold conditions with ADDR_EXPR early.

            gcc/testsuite/
            * gcc.dg/pr109546.c: New.

[Bug tree-optimization/109546] [13/14 Regression] Missed Dead Code Elimination when using __builtin_unreachable since r13-3596-ge7310e24b1c0ca

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109546

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|13.0                        |13.2

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 13.1 is being released, retargeting bugs to GCC 13.2.

[Bug tree-optimization/109546] [13/14 Regression] Missed Dead Code Elimination when using __builtin_unreachable since r13-3596-ge7310e24b1c0ca

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109546

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|13.2                        |13.3

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 13.2 is being released, retargeting bugs to GCC 13.3.

[Bug tree-optimization/109546] [13/14 Regression] Missed Dead Code Elimination when using __builtin_unreachable since r13-3596-ge7310e24b1c0ca

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109546

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
   Target Milestone|13.3                        |14.0
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #8 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed.