[Bug tree-optimization/102951] New: failure to optimize MIN_EXPR of subobject addresses of the same object

[Bug tree-optimization/102951] New: failure to optimize MIN_EXPR of subobject addresses of the same object

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102951

            Bug ID: 102951
           Summary: failure to optimize MIN_EXPR of subobject addresses of
                    the same object
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: msebor at gcc dot gnu.org
  Target Milestone: ---

GCC fails to fold the MIN_EXPR below to the lower address.  The test case
affects C because the C front end emits a MIN_EXPR but not C++ because the C++
front end emits an if statement.

$ cat z.c && gcc -O2 -S -Wall -fdump-tree-optimized=/dev/stdout z.c
extern int a[];

void f (void)
{
  int *p1 = &a[1];
  int *p2 = &a[2];

  int *d = p1 < p2 ? p1 : p2;
  *d = 0;
}

;; Function f (f, funcdef_no=0, decl_uid=1979, cgraph_uid=1, symbol_order=0)

void f ()
{
  int * d;

  <bb 2> [local count: 1073741824]:
  d_1 = MIN_EXPR <&a[2], &a[1]>;   // the two statements can be folded into
  *d_1 = 0;                        //   MEM[(int *)&a + 4B] = 0
  return;

}

[Bug tree-optimization/102951] failure to optimize MIN_EXPR of subobject addresses of the same object

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102951

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
extern int a[];

int *
foo (void)
{
  int *p1 = &a[1];
  int *p2 = &a[2];
  return p1 < p2 ? p1 : p2;
}

int
bar (void)
{
  int *p1 = &a[1];
  int *p2 = &a[2];
  return p1 < p2;
}

For the latter function, we optimize it in match.pd:
/* When the addresses are not directly of decls compare base and offset.
   This implements some remaining parts of fold_comparison address
   comparisons but still no complete part of it.  Still it is good
   enough to make fold_stmt not regress when not dispatching to fold_binary. 
*/
(for cmp (simple_comparison)
 (simplify
  (cmp (convert1?@2 addr@0) (convert2? addr@1))
  (with
   {
     poly_int64 off0, off1;
...

So, I guess for MIN_EXPR/MAX_EXPR with ADDR_EXPR operands, we can optimize it
similarly, the question is if we should try to do that through
repeating that huge code from there, or try to outline big parts of that into a
helper function, or perhaps could we e.g. do
  (with
    {
#if GENERIC
      tree l = generic_simplify (..., LT_EXPR, ...);
#else
      tree l = gimple_simplify (..., LT_EXPR, ...);
#endif
    }
    (if (l && integer_zerop (l))
      @0)
     (if (l && integer_nonzerop (l))
       @1)))
or so?
and therefore try to fold LT_EXPR instead of MIN_EXPR or MAX_EXPR and if that
folds into integer_zerop or integer_nonzerop

[Bug tree-optimization/102951] failure to optimize MIN_EXPR of subobject addresses of the same object

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102951

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2021-10-27
                 CC|                            |pinskia at gcc dot gnu.org
             Status|UNCONFIRMED                 |NEW

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed, this is funny because clang does not do it either.

[Bug tree-optimization/102951] failure to optimize MIN_EXPR of subobject addresses of the same object

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102951

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Factoring this somehow would be nicer, but not sure if the result will be
"nice" ...

[Bug tree-optimization/102951] failure to optimize MIN_EXPR of subobject addresses of the same object

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102951

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 51674
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51674&action=edit
gcc12-pr102951.patch

Untested fix.

[Bug tree-optimization/102951] failure to optimize MIN_EXPR of subobject addresses of the same object

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102951

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

https://gcc.gnu.org/g:6123b998b185572abac7d7224b34f03955bb91a2

commit r12-4776-g6123b998b185572abac7d7224b34f03955bb91a2
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Oct 28 20:10:15 2021 +0200

    match.pd: Optimize MIN_EXPR <addr1, addr2> etc. addr1 < addr2 would be
simplified [PR102951]

    This patch outlines the decision whether address comparison can be folded
    or not from the match.pd simple comparison simplification and uses it
    both there and in a new minmax simplification, such that we fold e.g.
    MAX (&a[2], &a[1]) etc.
    Some of the Wstringop-overflow-62.c changes might look weird, but that
    seems to be mainly due to gimple_fold_builtin_memset not bothering to
    copy over location, will fix that incrementally.

    2021-10-28  Jakub Jelinek  <jakub@redhat.com>

            PR tree-optimization/102951
            * fold-const.h (address_compare): Declare.
            * fold-const.c (address_compare): New function.
            * match.pd (cmp (convert1?@2 addr@0) (convert2? addr@1)): Use
            address_compare helper.
            (minmax cmp (convert1?@2 addr@0) (convert2?@3 addr@1)): New
            simplification.

            * gcc.dg/tree-ssa/pr102951.c: New test.
            * gcc.dg/Wstringop-overflow-62.c: Adjust expected diagnostics.