[Bug tree-optimization/110539] New: [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[Bug tree-optimization/110539] New: [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[theodort at inf dot ethz.ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

            Bug ID: 110539
           Summary: [14 Regression] Dead Code Elimination Regression at
                    since r14-338-g1dd154f6407
           Product: gcc
           Version: 14.0
            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: ---

https://godbolt.org/z/h9vrrj5je

Given the following code:

void foo(void);
static int a, c = 1;
static short b;
static int *d = &c, *e = &a;
static int **f = &d;
void __assert_fail() __attribute__((__noreturn__));
static void g(short h) {
    if (*d)
        ;
    else {
        if (e) __assert_fail();
        if (a) {
            __builtin_unreachable();
        } else
            __assert_fail();
    }
    if ((((0, 0) || h) == h) + b) *f = 0;
}
int main() {
    int i = 0 != 10 & a;
    g(i);
    *e = 9;
    e = 0;
    if (d == 0)
        ;
    else
        foo();
    ;
}

gcc-trunk -O2 does not eliminate the call to foo:

main:
        subq    $8, %rsp
        movq    d(%rip), %rax
        movq    e(%rip), %rdx
        movl    (%rax), %edi
        testl   %edi, %edi
        jne     .L2
        xorl    %eax, %eax
        call    __assert_fail
.L2:
        xorl    %ecx, %ecx
        xorl    %esi, %esi
        movq    %rcx, d(%rip)
        cmpq    $0, d(%rip)
        movl    $9, (%rdx)
        movq    %rsi, e(%rip)
        je      .L3
        call    foo
.L3:
        xorl    %eax, %eax
        popq    %rdx
        ret

gcc-13.1.0 -O2 eliminates the call to foo:

main:
        movq    d(%rip), %rax
        movq    e(%rip), %rdx
        movl    (%rax), %esi
        testl   %esi, %esi
        jne     .L2
        pushq   %rcx
        xorl    %eax, %eax
        call    __assert_fail
.L2:
        xorl    %eax, %eax
        movq    %rax, d(%rip)
        xorl    %eax, %eax
        movl    $9, (%rdx)
        xorl    %edx, %edx
        movq    %rdx, e(%rip)
        ret

Bisects to r14-338-g1dd154f6407

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |14.0

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1
           Assignee|unassigned at gcc dot gnu.org      |pinskia at gcc dot gnu.org
   Last reconfirmed|                            |2023-07-06

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---

  # RANGE [irange] int [0, 1] NONZERO 0x1
  i_7 = a.0_1 & 1;

  _17 = i_7 != 0;
  _12 = (int) _17;
  if (i_7 == _12)


So this should have been optimized to _17 = (bool) i_7;
and then if (1)

Maybe it is an order of the stuff in match.pd ...

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
So the difference comes from the order. Before in phiopt we had:
-  /* Defer boolean x ? 0 : {1,-1} or x ? {1,-1} : 0 to
-     match_simplify_replacement.  */
-  if (TREE_CODE (TREE_TYPE (lhs)) == BOOLEAN_TYPE
-      && (integer_zerop (arg0)
-         || integer_zerop (arg1)
-         || TREE_CODE (TREE_TYPE (arg0)) == BOOLEAN_TYPE
-         || (TYPE_PRECISION (TREE_TYPE (arg0))
-             <= TYPE_PRECISION (TREE_TYPE (lhs)))))
-    return false;

But now the order is such that `?0:{1,-1}`, `?{1,-1}:0` is handled first.

So what we need to pattern match here is `(convert)zero_one_value_p@0!=0` and
simplify that into just (convert)@0

(for neeq (ne eq)
 (simplify
  (convert (neeq zero_one_value_p@0 integer_zerop))
  (if (neeq == NE_EXPR))
   (convert @0)
   (bit_xor (convert @0) { build_one_cst (type); } ))

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Here is a testcase for the missing optimization (at -O1) which is optimized at
the RTL level (for some targets but not all):
```
int f(int a)
{
        int b = a & 1;
        int c = b != 0;
        return c == b;
}
```

Though it does optimize at -O2 because VRP changes: `b != 0` into `(_Bool)
b_4`.

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
yes adding:
/* (convert)(zeroone != 0) into (convert)zeroone */
/* (convert)(zeroone == 0) into ((convert)zeroone)^1 */
(for neeq (ne eq)
 (simplify
  (convert (neeq zero_one_valued_p@0 integer_zerop))
  (if (neeq == NE_EXPR)
   (convert @0)
   (bit_xor (convert @0) { build_one_cst (type); } ))))


Fixes the original testcase.

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #4)
> yes adding:
> /* (convert)(zeroone != 0) into (convert)zeroone */
> /* (convert)(zeroone == 0) into ((convert)zeroone)^1 */
> (for neeq (ne eq)
>  (simplify
>   (convert (neeq zero_one_valued_p@0 integer_zerop))
>   (if (neeq == NE_EXPR)
>    (convert @0)
>    (bit_xor (convert @0) { build_one_cst (type); } ))))
> 
> 
> Fixes the original testcase.

One simple regression:
/* { dg-final { scan-tree-dump-times "Folded to: _\[0-9\]+ = 1;" 1 "dom3"} } */
/* { dg-final { scan-tree-dump-times "Folded to: _\[0-9\]+ = 0;" 1 "dom3"} } */

fail now but that is they just don't match the above on accident.

Before:
  Replaced 'bufferstep_36' with constant '0'
gimple_simplified to _5 = 1;
  Folded to: _5 = 1;
After:
  Replaced 'bufferstep_36' with constant '0'
gimple_simplified to bufferstep_23 = 1;
  Folded to: bufferstep_23 = 1;

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                URL|                            |https://gcc.gnu.org/piperma
                   |                            |il/gcc-patches/2023-July/62
                   |                            |3881.html
           Keywords|                            |patch

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Patch posted:
https://gcc.gnu.org/pipermail/gcc-patches/2023-July/623881.html

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Depends on|                            |110293

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #1)
>   # RANGE [irange] int [0, 1] NONZERO 0x1
>   i_7 = a.0_1 & 1;
> 
>   _17 = i_7 != 0;
>   _12 = (int) _17;
>   if (i_7 == _12)

Wait that is:
i_7 == (convert) (i_7 != 0);

Which is basically PR 110293. So there is another way of fixing this is
simplifying `i_7 == (convert) (i_7 != 0)` into  `(unsigned_type)i_7 <= 1` and
that is always true as `(a.0_1 & 1) <= 1` is always true.

Let me try fixing PR 110293 as the way of fixing this one instead of the patch
which I submitted but has other issues.


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110293
[Bug 110293] Some `A CMP (A NEEQ 0)` is not simplified in some cases

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

--- Comment #8 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Let me file a different bug for onezero stuff that I tried out ...

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                URL|https://gcc.gnu.org/piperma |https://gcc.gnu.org/piperma
                   |il/gcc-patches/2023-July/62 |il/gcc-patches/2023-July/62
                   |3881.html                   |4293.html

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
New patch submitted:
https://gcc.gnu.org/pipermail/gcc-patches/2023-July/624293.html

Which also gets rid of the call to foo at -O1 -fstrict-aliasing too (which was
not done in GCC 13 either).

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

--- Comment #10 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:285c9d042e90a7425b37697edc9ec93a1b03b486

commit r14-2501-g285c9d042e90a7425b37697edc9ec93a1b03b486
Author: Andrew Pinski <apinski@marvell.com>
Date:   Wed Jul 12 00:33:14 2023 -0700

    Fix part of PR 110293: `A NEEQ (A NEEQ CST)` part

    This fixes part of PR 110293, for the outer comparison case
    being `!=` or `==`.  In turn PR 110539 is able to be optimized
    again as the if statement for `(a&1) == ((a & 1) != 0)` gets optimized
    to `false` early enough to allow FRE/DOM to do a CSE for memory store/load.

    OK? Bootstrapped and tested on x86_64-linux with no regressions.

    gcc/ChangeLog:

            PR tree-optimization/110293
            PR tree-optimization/110539
            * match.pd: Expand the `x != (typeof x)(x == 0)`
            pattern to handle where the inner and outer comparsions
            are either `!=` or `==` and handle other constants
            than 0.

    gcc/testsuite/ChangeLog:

            * gcc.dg/tree-ssa/pr110293-1.c: New test.
            * gcc.dg/tree-ssa/pr110539-1.c: New test.
            * gcc.dg/tree-ssa/pr110539-2.c: New test.
            * gcc.dg/tree-ssa/pr110539-3.c: New test.
            * gcc.dg/tree-ssa/pr110539-4.c: New test.

[Bug tree-optimization/110539] [14 Regression] Dead Code Elimination Regression at since r14-338-g1dd154f6407

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110539

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

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

--- Comment #11 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed.