[Bug tree-optimization/110134] New: [10/11/12/13/14 Regression] (-unsigned1) != CST is not optimized to unsigned1 != CST at the gimple level

[Bug tree-optimization/110134] New: [10/11/12/13/14 Regression] (-unsigned1) != CST is not optimized to unsigned1 != CST at the gimple level

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110134

            Bug ID: 110134
           Summary: [10/11/12/13/14 Regression] (-unsigned1) != CST is not
                    optimized to unsigned1 != CST at the gimple level
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: missed-optimization, TREE
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---

Take:
```
bool f1(int  a)
{
  unsigned t = a < 10;
  t = -t;
  return t != 0;
}

bool f2(int  a)
{
  signed t = a < 10;
  t = -t;
  return t != 0;
}
```

These should both optimize to just `a >= 9` in forwprop1 .
Both of these were working in GCC 5.
Starting GCC 6, neither was done and then in GCC 8, f2 was handled.

I noticed this while looking at PR 110131 (doing this does not fix that
though).

[Bug tree-optimization/110134] [10/11/12/13/14 Regression] (-unsigned1) != CST is not optimized to unsigned1 != CST at the gimple level

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110134

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2023-06-06
             Status|UNCONFIRMED                 |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |pinskia at gcc dot gnu.org

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I have a simple patch.
TYPE_OVERFLOW_UNDEFINED (TREE_TYPE (@0))
Can just be:
(cmp == EQ_EXPR || cmp == NE_EXPR || TYPE_OVERFLOW_UNDEFINED (TREE_TYPE (@0)))

For /* -A CMP -B -> B CMP A.  */

Because -a == -b is still a == b, no matter what.

[Bug tree-optimization/110134] [10/11/12/13/14 Regression] (-unsigned1) != CST is not optimized to unsigned1 != CST at the gimple level

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110134

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
That is:
```
/* -A CMP -B -> B CMP A.  */
(for cmp (tcc_comparison)
     scmp (swapped_tcc_comparison)
 (simplify
  (cmp (negate @0) (negate @1))
  (if (FLOAT_TYPE_P (TREE_TYPE (@0))
       || (ANY_INTEGRAL_TYPE_P (TREE_TYPE (@0))
           && (cmp == EQ_EXPR
               || cmp == NE_EXPR
               || TYPE_OVERFLOW_UNDEFINED (TREE_TYPE (@0)))))
   (scmp @0 @1)))
 (simplify
  (cmp (negate @0) CONSTANT_CLASS_P@1)
  (if (FLOAT_TYPE_P (TREE_TYPE (@0))
       || (ANY_INTEGRAL_TYPE_P (TREE_TYPE (@0))
           && (cmp == EQ_EXPR
               || cmp == NE_EXPR
               || TYPE_OVERFLOW_UNDEFINED (TREE_TYPE (@0)))))
   (with { tree tem = const_unop (NEGATE_EXPR, TREE_TYPE (@0), @1); }
    (if (tem && !TREE_OVERFLOW (tem))
     (scmp @0 { tem; }))))))
```

[Bug tree-optimization/110134] [10/11/12/13/14 Regression] (-unsigned1) != CST is not optimized to unsigned1 != CST at the gimple level

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110134

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
r6-1814-g66e1cacf608045 caused GCC 6 to stop doing both.
Because they were considered redundant with the patterns added by:
r6-1113-g534bd33b61d08e

Which was mostly true.

I think forwprop was working again with r8-2194-g4f450a2b102212 for f2 (I am
not 100% sure).

So to summarize r6-1814-g66e1cacf608045 removed the unsigned/wrapping case for
NE/EQ since it was originally thought as redundant. My patch adds support back
for those 2.

case EQ_EXPR:
case NE_EXPR:
...

-      /* Similarly for a NEGATE_EXPR.  */
-      if (TREE_CODE (arg0) == NEGATE_EXPR
-         && TREE_CODE (arg1) == INTEGER_CST
-         && 0 != (tem = negate_expr (fold_convert_loc (loc, TREE_TYPE (arg0),
-                                                       arg1)))
-         && TREE_CODE (tem) == INTEGER_CST
-         && !TREE_OVERFLOW (tem))
-       return fold_build2_loc (loc, code, type, TREE_OPERAND (arg0, 0), tem);

...

-      /* Fold -X op -Y as X op Y, where op is eq/ne.  */
-      if (TREE_CODE (arg0) == NEGATE_EXPR
-          && TREE_CODE (arg1) == NEGATE_EXPR)
-       return fold_build2_loc (loc, code, type,
-                               TREE_OPERAND (arg0, 0),
-                               fold_convert_loc (loc, TREE_TYPE (arg0),
-                                                 TREE_OPERAND (arg1, 0)));


Since these 2 did cases didn't test TYPE_OVERFLOW_UNDEFINED, we should handle
EQ/NE also in match.pd the same way.

[Bug tree-optimization/110134] [10/11/12/13/14 Regression] (-unsigned1) != CST is not optimized to unsigned1 != CST at the gimple level

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110134

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2
   Target Milestone|---                         |10.5

[Bug tree-optimization/110134] [10/11/12/13/14 Regression] (-unsigned1) != CST is not optimized to unsigned1 != CST at the gimple level

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110134

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                URL|                            |https://gcc.gnu.org/piperma
                   |                            |il/gcc-patches/2023-June/62
                   |                            |0818.html
           Keywords|                            |patch

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Patch submitted with a few testcases to make sure it will not regress again:
https://gcc.gnu.org/pipermail/gcc-patches/2023-June/620818.html

[Bug tree-optimization/110134] [10/11/12/13/14 Regression] (-unsigned1) != CST is not optimized to unsigned1 != CST at the gimple level

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110134

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

https://gcc.gnu.org/g:cc155ff9c38848a8e6a7125dd0b66ac0aef47880

commit r14-1599-gcc155ff9c38848a8e6a7125dd0b66ac0aef47880
Author: Andrew Pinski <apinski@marvell.com>
Date:   Mon Jun 5 19:12:43 2023 -0700

    For the `-A CMP -B -> B CMP A` pattern allow EQ/NE for all integer types

    I noticed while looking at some code generation issue, that forwprop
    was not handling `-a == 0` for unsigned types and I was confused why
    it was not.
    r6-1814-g66e1cacf608045 removed these from fold because they
    were supposed to be already handled by the match.pd patterns
    but it was missed that the match.pd patterns checked
    TYPE_OVERFLOW_UNDEFINED while fold didn't do that for NE/EQ.
    This patch removes the restriction on NE/EQ on TYPE_OVERFLOW_UNDEFINED.

    OK? Bootstrapped and tested on x86_64-linux-gnu.

    gcc/ChangeLog:

            PR tree-optimization/110134
            * match.pd (-A CMP -B -> B CMP A): Allow EQ/NE for all integer
            types.
            (-A CMP CST -> B CMP (-CST)): Likewise.

    gcc/testsuite/ChangeLog:

            PR tree-optimization/110134
            * gcc.dg/tree-ssa/negneq-1.c: New test.
            * gcc.dg/tree-ssa/negneq-2.c: New test.
            * gcc.dg/tree-ssa/negneq-3.c: New test.
            * gcc.dg/tree-ssa/negneq-4.c: New test.

[Bug tree-optimization/110134] [10/11/12/13/14 Regression] (-unsigned1) != CST is not optimized to unsigned1 != CST at the gimple level

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110134

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
   Target Milestone|10.5                        |14.0
         Resolution|---                         |FIXED

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed on the trunk for GCC 14, the performance regression due to this bug I
doubt is noticable since it had not been noticed since GCC 6 even (well PR
14753 noticed it but it was similar to this one, it was not noticed in real
code yet).