[Bug target/109502] New: [12/13 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[Bug target/109502] New: [12/13 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[zsojka at seznam dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

            Bug ID: 109502
           Summary: [12/13 Regression] wrong code with -O -ftree-vectorize
                    -fvect-cost-model=unlimited on aarch64
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: zsojka at seznam dot cz
  Target Milestone: ---
              Host: x86_64-pc-linux-gnu
            Target: aarch64-unknown-linux-gnu

Created attachment 54853
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54853&action=edit
reduced testcase

Output:
$ aarch64-unknown-linux-gnu-gcc -O -ftree-vectorize -fvect-cost-model=unlimited
testcase.c -static
$ qemu-aarch64 -- ./a.out 
qemu: uncaught target signal 6 (Aborted) - core dumped
Aborted


The value is 0xffffffff instead of 1.

$ aarch64-unknown-linux-gnu-gcc -v
Using built-in specs.
COLLECT_GCC=/repo/gcc-trunk/binary-latest-aarch64/bin/aarch64-unknown-linux-gnu-gcc
COLLECT_LTO_WRAPPER=/repo/gcc-trunk/binary-trunk-r13-7165-20230413001648-g66c7257b675-checking-yes-rtl-df-extra-aarch64/bin/../libexec/gcc/aarch64-unknown-linux-gnu/13.0.1/lto-wrapper
Target: aarch64-unknown-linux-gnu
Configured with: /repo/gcc-trunk//configure --enable-languages=c,c++
--enable-valgrind-annotations --disable-nls --enable-checking=yes,rtl,df,extra
--with-cloog --with-ppl --with-isl
--with-sysroot=/usr/aarch64-unknown-linux-gnu --build=x86_64-pc-linux-gnu
--host=x86_64-pc-linux-gnu --target=aarch64-unknown-linux-gnu
--with-ld=/usr/bin/aarch64-unknown-linux-gnu-ld
--with-as=/usr/bin/aarch64-unknown-linux-gnu-as --disable-libstdcxx-pch
--prefix=/repo/gcc-trunk//binary-trunk-r13-7165-20230413001648-g66c7257b675-checking-yes-rtl-df-extra-aarch64
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 13.0.1 20230413 (experimental) (GCC)

[Bug target/109502] [12/13 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |12.3

[Bug tree-optimization/109502] [12/13 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |tree-optimization
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2023-04-14

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

  g.0_1 = g;
  _2 = g.0_1 == 0;
  a_7 = (unsigned int) _2;
  _3 = a_7 % 6;
  _4 = _3 == 0;
  _5 = (unsigned int) _4;
  a_8 = _5 + a_7;

To:

  g.0_1 = g;
  _2 = g.0_1 == 0;
  a_7 = (unsigned int) _2;
  _3 = a_7 % 6;
  _15 = {_3, g.0_1};
  mask__4.4_16 = { 0, 0 } == _15;
  vect__5.5_19 = VIEW_CONVERT_EXPR<vector(2) unsigned int>(mask__4.4_16);
  _17 = BIT_FIELD_REF <mask__4.4_16, 32, 0>;
  _18 = (bool) _17;
  _4 = _3 == 0;
  _5 = (unsigned int) _18;
  _20 = .REDUC_PLUS (vect__5.5_19);
  a_8 = _20;

Which is wrong. The if anything there is a missing - missing after the
reduc_plus (or before) when it translates the bools comparisons into vector
comparisons.

[Bug tree-optimization/109502] [12/13 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #1)
> SLP transforms:
> 
>   g.0_1 = g;
>   _2 = g.0_1 == 0;
>   a_7 = (unsigned int) _2;
>   _3 = a_7 % 6;
>   _4 = _3 == 0;
>   _5 = (unsigned int) _4;
>   a_8 = _5 + a_7;
> 
> To:
> 
>   g.0_1 = g;
>   _2 = g.0_1 == 0;
>   a_7 = (unsigned int) _2;
>   _3 = a_7 % 6;
>   _15 = {_3, g.0_1};
>   mask__4.4_16 = { 0, 0 } == _15;
>   vect__5.5_19 = VIEW_CONVERT_EXPR<vector(2) unsigned int>(mask__4.4_16);
>   _17 = BIT_FIELD_REF <mask__4.4_16, 32, 0>;
>   _18 = (bool) _17;
>   _4 = _3 == 0;
>   _5 = (unsigned int) _18;
>   _20 = .REDUC_PLUS (vect__5.5_19);
>   a_8 = _20;
> 

If anything there is a missing, a negative after the
reduc_plus (or before) when it translates the bools comparisons into vector
comparisons.

[Bug tree-optimization/109502] [12/13 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #2)
> (In reply to Andrew Pinski from comment #1)
> > SLP transforms:
> > 
> >   g.0_1 = g;
> >   _2 = g.0_1 == 0;
> >   a_7 = (unsigned int) _2;
> >   _3 = a_7 % 6;
> >   _4 = _3 == 0;
> >   _5 = (unsigned int) _4;
> >   a_8 = _5 + a_7;
> > 
> > To:
> > 
> >   g.0_1 = g;
> >   _2 = g.0_1 == 0;
> >   a_7 = (unsigned int) _2;
> >   _3 = a_7 % 6;
> >   _15 = {_3, g.0_1};
> >   mask__4.4_16 = { 0, 0 } == _15;
> >   vect__5.5_19 = VIEW_CONVERT_EXPR<vector(2) unsigned int>(mask__4.4_16);
> >   _17 = BIT_FIELD_REF <mask__4.4_16, 32, 0>;
> >   _18 = (bool) _17;
> >   _4 = _3 == 0;
> >   _5 = (unsigned int) _18;
> >   _20 = .REDUC_PLUS (vect__5.5_19);
> >   a_8 = _20;
> > 
> 
> If anything there is a missing, a negative after the
> reduc_plus (or before) when it translates the bools comparisons into vector
> comparisons.

Indeed.  Usually this is the failure of bool pattern detection.

unsigned foo (unsigned *p)
{
  unsigned tem1 = p[0] == 0;
  unsigned tem2 = p[1] == 0;
  unsigned tem3 = p[2] == 0;
  unsigned tem4 = p[3] == 0;
  return tem1 + tem2 + tem3 + tem4;
}

doesn't reproduce it - we have put in defences "after the fact" to work
around this for some cases:

t.c:7:29: note:   ==> examining statement: tem4_16 = (unsigned int) _8;
t.c:7:29: note:   vect_is_simple_use: operand _7 == 0, type of def: internal
t.c:7:29: missed:   type conversion to/from bit-precision unsupported.
t.c:7:29: note:   vect_is_simple_use: operand _7 == 0, type of def: internal
t.c:7:29: missed:   mixed mask and nonmask vector types

[Bug tree-optimization/109502] [12/13 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

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

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

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Simple omission.

[Bug tree-optimization/109502] [12/13 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

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

https://gcc.gnu.org/g:bf24f2db2841b97bc5e86bf9294d61eef32f83b3

commit r13-7180-gbf24f2db2841b97bc5e86bf9294d61eef32f83b3
Author: Richard Biener <rguenther@suse.de>
Date:   Fri Apr 14 09:55:27 2023 +0200

    tree-optimization/109502 - vector conversion between mask and non-mask

    The following fixes a check that should have rejected vectorizing
    a conversion between a mask and non-mask type.  Those should be
    done via pattern statements.

            PR tree-optimization/109502
            * tree-vect-stmts.cc (vectorizable_assignment): Fix
            check for conversion between mask and non-mask types.

            * gcc.dg/vect/pr109502.c: New testcase.

[Bug tree-optimization/109502] [12 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[12/13 Regression] wrong    |[12 Regression] wrong code
                   |code with -O                |with -O -ftree-vectorize
                   |-ftree-vectorize            |-fvect-cost-model=unlimited
                   |-fvect-cost-model=unlimited |on aarch64
                   |on aarch64                  |
           Priority|P3                          |P2
      Known to fail|13.0                        |
      Known to work|                            |13.0

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Should be fixed on trunk.  Is probably latent on older branches as well.

[Bug tree-optimization/109502] [12 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

--- Comment #7 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Richard Biener
<rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:08605396d164c57eb3c80051001b86b96da0405c

commit r12-9414-g08605396d164c57eb3c80051001b86b96da0405c
Author: Richard Biener <rguenther@suse.de>
Date:   Fri Apr 14 09:55:27 2023 +0200

    tree-optimization/109502 - vector conversion between mask and non-mask

    The following fixes a check that should have rejected vectorizing
    a conversion between a mask and non-mask type.  Those should be
    done via pattern statements.

            PR tree-optimization/109502
            * tree-vect-stmts.cc (vectorizable_assignment): Fix
            check for conversion between mask and non-mask types.

            * gcc.dg/vect/pr109502.c: New testcase.

    (cherry picked from commit bf24f2db2841b97bc5e86bf9294d61eef32f83b3)

[Bug tree-optimization/109502] [11 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|12.3                        |11.4
      Known to work|                            |12.2.1
            Summary|[12 Regression] wrong code  |[11 Regression] wrong code
                   |with -O -ftree-vectorize    |with -O -ftree-vectorize
                   |-fvect-cost-model=unlimited |-fvect-cost-model=unlimited
                   |on aarch64                  |on aarch64
      Known to fail|12.2.1                      |12.2.0

--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
Also picks to 11.

[Bug tree-optimization/109502] [11 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

--- Comment #9 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Richard Biener
<rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:7fc59677f7a9a2bff75e0002fef303a41a9ebe9d

commit r11-10677-g7fc59677f7a9a2bff75e0002fef303a41a9ebe9d
Author: Richard Biener <rguenther@suse.de>
Date:   Fri Apr 14 09:55:27 2023 +0200

    tree-optimization/109502 - vector conversion between mask and non-mask

    The following fixes a check that should have rejected vectorizing
    a conversion between a mask and non-mask type.  Those should be
    done via pattern statements.

            PR tree-optimization/109502
            * tree-vect-stmts.c (vectorizable_assignment): Fix
            check for conversion between mask and non-mask types.

            * gcc.dg/vect/pr109502.c: New testcase.

    (cherry picked from commit bf24f2db2841b97bc5e86bf9294d61eef32f83b3)

[Bug tree-optimization/109502] [11 Regression] wrong code with -O -ftree-vectorize -fvect-cost-model=unlimited on aarch64

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109502

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED
      Known to fail|                            |11.3.0

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.