[Bug tree-optimization/113433] New: [12/13/14 Regression] Missed optimization for redundancy computation elimination

[Bug tree-optimization/113433] New: [12/13/14 Regression] Missed optimization for redundancy computation elimination

[652023330028 at smail dot nju.edu.cn]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113433

            Bug ID: 113433
           Summary: [12/13/14 Regression] Missed optimization for
                    redundancy computation elimination
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: 652023330028 at smail dot nju.edu.cn
  Target Milestone: ---

Hello, we noticed that maybe there is a missed optimization for redundancy
computation elimination.

Different from PR 113265, This code worked as expected until GCC-12.

https://godbolt.org/z/nKh7zfsYn
int c,d,e;
void func(int y){
    c=1-y;
    d=0;
    e=c/(d+(-c));
}

GCC (trunk):
func(int):
        mov     DWORD PTR d[rip], 0
        mov     eax, 1
        sub     eax, edi
        sub     edi, 1
        cdq
        mov     DWORD PTR c[rip], eax
        idiv    edi
        mov     DWORD PTR e[rip], eax
        ret

Expected code (GCC-11.4):
func(int):
        mov     DWORD PTR d[rip], 0
        mov     eax, 1
        mov     DWORD PTR e[rip], -1
        sub     eax, edi
        mov     DWORD PTR c[rip], eax
        ret

Thank you very much for your time and effort! We look forward to hearing from
you.

[Bug tree-optimization/113433] [12/13/14 Regression] Missed optimization for redundancy computation elimination

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113433

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2024-01-17
           Assignee|unassigned at gcc dot gnu.org      |pinskia at gcc dot gnu.org

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
But it is a similar issue, we are left with:
  _1 = 1 - y_7(D);
  _5 = y_7(D) + -1;
  _6 = _1 / _5;

Which is not folded as being as we don't know that `1 - x` and `x + -1` are
negative opposites.

[Bug tree-optimization/113433] [12/13/14 Regression] Missed optimization for redundancy computation elimination

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113433

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2
   Target Milestone|---                         |12.4

[Bug tree-optimization/113433] [12/13/14 Regression] Missed optimization for redundancy computation elimination

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113433

Jeffrey A. Law <law at gcc dot gnu.org> changed:

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

--- Comment #2 from Jeffrey A. Law <law at gcc dot gnu.org> ---
So we could attack this as a DOM problem.  Not all the infrastructure is in
there.  But the recognition of negated expressions isn't hard.  Something like
this in tree-ssa-scopedtables will detect the negated expression in the hash
table.  

            /* We might be able to lookup the negated expression.  */
            case PLUS_EXPR:
              {
                tree x = gimple_assign_rhs_to_tree (stmt);
                x = fold_unary (NEGATE_EXPR, TREE_TYPE (x), x);
                struct hashable_expr expr;
                expr.type = TREE_TYPE (x);
                expr.kind = EXPR_BINARY;
                expr.ops.binary.op = MINUS_EXPR;
                expr.ops.binary.opnd0 = TREE_OPERAND (x, 0);
                expr.ops.binary.opnd1 = TREE_OPERAND (x, 1);
                class expr_hash_elt element2 (&expr, NULL_TREE);
                expr_hash_elt **slot
                  = m_avail_exprs->find_slot (&element2, NO_INSERT);
                if (slot && *slot)
                  return fold_build1 (NEGATE_EXPR, TREE_TYPE (x),  (*slot)->lhs
());
                return NULL_TREE;
              }

Right now DOM isn't prepared for avail_expr_stack::simplify_binary_operation to
return anything other than a constant, ssa_name or NULL.  But how hard could it
be to expand further :-)


Not sure if this happens enough to make the extra lookups worth the effort.

[Bug tree-optimization/113433] [12/13/14/15 Regression] Missed optimization for redundancy computation elimination

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113433

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Created attachment 58288
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=58288&action=edit
Patch which works

Just need to add testcases.

[Bug tree-optimization/113433] [12/13/14/15 Regression] Missed optimization for redundancy computation elimination

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113433

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

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

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Dup of bug 113265 really. No reason to keep both open.

*** This bug has been marked as a duplicate of bug 113265 ***