[Bug target/103661] New: __builtin_cpu_supports returns a negative integer for avx512vbmi2

[Bug target/103661] New: __builtin_cpu_supports returns a negative integer for avx512vbmi2

[stefan.kneifel at bluewin dot ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

            Bug ID: 103661
           Summary: __builtin_cpu_supports returns a negative integer for
                    avx512vbmi2
           Product: gcc
           Version: 11.2.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: stefan.kneifel at bluewin dot ch
  Target Milestone: ---

__builtin_cpu_supports is supposed to return a positive integer if the run-time
CPU supports the requested feature and zero otherwise (see Documentation).
However, for the feature "avx512vbmi2", a negative integer (0x80000000) is
returned.

FEATURE_AVX512VBMI2 equals to 31 in gcc/common/config/i386/i386-cpuinfo.h, thus
the returned 32-bit integer has bit 31 set and is therefore negative.

This might lead to surprising errors even inside GCC (e.g. when features are
tested in automatically generated resolvers; the code seems to rely on the
positivity).

Would it be an acceptable solution to declare __builtin_cpu_supports as
unsigned, or do you see another solution?

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Oh it does say positive integer:
This function returns a positive integer if the run-time CPU supports feature
and returns 0 otherwise.

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64
           Keywords|                            |wrong-code

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
https://gcc.gnu.org/onlinedocs/gcc-11.2.0/gcc/x86-Built-in-Functions.html#index-_005f_005fbuiltin_005fcpu_005fsupports-1

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[stefan.kneifel at bluewin dot ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

--- Comment #3 from Stefan Kneifel <stefan.kneifel at bluewin dot ch> ---
Created attachment 51975
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51975&action=edit
First try of a patch, making it unsigned (and correcting the resolver
generator)

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[stefan.kneifel at bluewin dot ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

--- Comment #4 from Stefan Kneifel <stefan.kneifel at bluewin dot ch> ---
Created attachment 51978
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51978&action=edit
Testcase

Without the proposed patch, it compiles to:
--------
0000000000000000 <has_vbmi>:
   0:   50                      push   %rax
   1:   e8 00 00 00 00          call   6 <has_vbmi+0x6>
                        2: R_X86_64_PLT32       __cpu_indicator_init-0x4
   6:   31 c0                   xor    %eax,%eax
   8:   f6 05 00 00 00 00 04    testb  $0x4,0x0(%rip)        # f <has_vbmi+0xf>
                        a: R_X86_64_PC32        __cpu_model+0xa
   f:   5a                      pop    %rdx
  10:   0f 95 c0                setne  %al
  13:   c3                      ret    

0000000000000014 <has_vbmi2>:
  14:   50                      push   %rax
  15:   e8 00 00 00 00          call   1a <has_vbmi2+0x6>
                        16: R_X86_64_PLT32      __cpu_indicator_init-0x4
  1a:   31 c0                   xor    %eax,%eax
  1c:   5a                      pop    %rdx
  1d:   c3                      ret    
--------
returning 1 only for avx512vbmi, but not for avx512vbmi2, where it returns
always zero regardless whether the machine has it or not.


With the proposed patch, it compiles to:
--------
0000000000000000 <has_vbmi>:
   0:   50                      push   %rax
   1:   e8 00 00 00 00          call   6 <has_vbmi+0x6>
                        2: R_X86_64_PLT32       __cpu_indicator_init-0x4
   6:   8b 05 00 00 00 00       mov    0x0(%rip),%eax        # c <has_vbmi+0xc>
                        8: R_X86_64_PC32        __cpu_model+0x8
   c:   5a                      pop    %rdx
   d:   c1 e8 1a                shr    $0x1a,%eax
  10:   83 e0 01                and    $0x1,%eax
  13:   c3                      ret    

0000000000000014 <has_vbmi2>:
  14:   50                      push   %rax
  15:   e8 00 00 00 00          call   1a <has_vbmi2+0x6>
                        16: R_X86_64_PLT32      __cpu_indicator_init-0x4
  1a:   8b 05 00 00 00 00       mov    0x0(%rip),%eax        # 20
<has_vbmi2+0xc>
                        1c: R_X86_64_PC32       __cpu_model+0x8
  20:   5a                      pop    %rdx
  21:   c1 e8 1f                shr    $0x1f,%eax
  24:   c3                      ret    
--------
returning correctly 1 if the machine has avx512vbmi2.

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

Martin Liška <marxin at gcc dot gnu.org> changed:

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

--- Comment #5 from Martin Liška <marxin at gcc dot gnu.org> ---
@Stefan: Great, can you please send the patch to the mailing list?

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

--- Comment #6 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Martin Liska <marxin@gcc.gnu.org>:

https://gcc.gnu.org/g:127c7178d5ec502d95862fd823537cbca1a0cb99

commit r12-5993-g127c7178d5ec502d95862fd823537cbca1a0cb99
Author: Martin Liska <mliska@suse.cz>
Date:   Mon Dec 13 15:34:30 2021 +0100

    i386: Fix emissing of __builtin_cpu_supports.

            PR target/103661

    gcc/ChangeLog:

            * config/i386/i386-builtins.c (fold_builtin_cpu): Compare to 0
            as API expects that non-zero values are returned (do that
            it mask == 31).
            For "avx512vbmi2" argument, we return now 1 << 31, which is a
            negative integer value.

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2021-12-15
      Known to work|                            |12.0

--- Comment #7 from Martin Liška <marxin at gcc dot gnu.org> ---
Fixed on master so far.

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

Martin Liška <marxin at gcc dot gnu.org> changed:

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

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

--- Comment #8 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Martin Liska
<marxin@gcc.gnu.org>:

https://gcc.gnu.org/g:43d67e5e75627ad1f5ba2e983a730f37547ba9b7

commit r11-9394-g43d67e5e75627ad1f5ba2e983a730f37547ba9b7
Author: Martin Liska <mliska@suse.cz>
Date:   Mon Dec 13 15:34:30 2021 +0100

    i386: Fix emissing of __builtin_cpu_supports.

            PR target/103661

    gcc/ChangeLog:

            * config/i386/i386-builtins.c (fold_builtin_cpu): Compare to 0
            as API expects that non-zero values are returned (do that
            it mask == 31).
            For "avx512vbmi2" argument, we return now 1 << 31, which is a
            negative integer value.

    (cherry picked from commit 127c7178d5ec502d95862fd823537cbca1a0cb99)

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

--- Comment #9 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-10 branch has been updated by Martin Liska
<marxin@gcc.gnu.org>:

https://gcc.gnu.org/g:a2af3581cba0e7fa662bd8bed7e3b90451ef6a94

commit r10-10341-ga2af3581cba0e7fa662bd8bed7e3b90451ef6a94
Author: Martin Liska <mliska@suse.cz>
Date:   Mon Dec 13 15:34:30 2021 +0100

    i386: Fix emissing of __builtin_cpu_supports.

            PR target/103661

    gcc/ChangeLog:

            * config/i386/i386-builtins.c (fold_builtin_cpu): Compare to 0
            as API expects that non-zero values are returned (do that
            it mask == 31).
            For "avx512vbmi2" argument, we return now 1 << 31, which is a
            negative integer value.

    (cherry picked from commit 127c7178d5ec502d95862fd823537cbca1a0cb99)

[Bug target/103661] __builtin_cpu_supports returns a negative integer for avx512vbmi2

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103661

Martin Liška <marxin at gcc dot gnu.org> changed:

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

--- Comment #10 from Martin Liška <marxin at gcc dot gnu.org> ---
Fixed in gcc-10,11 braches.