[Bug tree-optimization/103325] New: 1 << -1 is never reduced to a constant during gimple

[Bug tree-optimization/103325] New: 1 << -1 is never reduced to a constant during gimple

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103325

            Bug ID: 103325
           Summary: 1 << -1 is never reduced to a constant during gimple
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---

Take:
int main() {
  return 1 >> (-1);
}

GCC is the only compiler I have tried (MSVC, ICC and clang/LLVM) which does not
remove the shift.
Yes this is undefined behavior but really I think it is best to reduce to
something rather than keeping around the shift. Even converting it to
__builtin_unreachable will be ok in my book.

[Bug tree-optimization/103325] 1 << -1 is never reduced to a constant during gimple

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103325

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I should note I noticed this while working on PR 103314.

[Bug tree-optimization/103325] 1 << -1 is never reduced to a constant during gimple

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103325

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2021-11-19
     Ever confirmed|0                           |1

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think we used to constant fold this as 1<<1.  GCC 7 does

> gcc-7 -S t.c -fdump-tree-original
t.c: In function 'main':
t.c:2:12: warning: right shift count is negative [-Wshift-count-negative]
   return 1 >> (-1);
            ^~
> cat t.c.003t.original 

;; Function main (null)
;; enabled by -tree-original


{
  return 2;
}
return 0;

but IIRC that behavior was removed from {int_const,wide}_int_binop at some
point, maybe also to enable sanitization.

gimple-ssa-isolate-paths.c would be one place to turn such code into
unreachable or traps (see other PRs to make the behavior configurable).

[Bug tree-optimization/103325] 1 << -1 is never reduced to a constant during gimple

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103325

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=96929

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #2)

> but IIRC that behavior was removed from {int_const,wide}_int_binop at some
> point, maybe also to enable sanitization.

This was only done in GCC 11 by  r11-5271-g4866b2f5db117f9 (PR96929).

[Bug tree-optimization/103325] 1 << -1 is never reduced to a constant during gimple

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103325

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
   Target Milestone|---                         |12.3
             Status|NEW                         |RESOLVED

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed for GCC 12.3.0 by r12-9091-g9dccaeaa586a16 (aka PR 108306 ). So closing
as fixed.

Note DOM and VRP both do it so it is done at -O1 and -O2.