[Bug target/101022] New: rs6000: __builtin_altivec_vcmpequt expands to wrong pattern

[Bug target/101022] New: rs6000: __builtin_altivec_vcmpequt expands to wrong pattern

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101022

            Bug ID: 101022
           Summary: rs6000: __builtin_altivec_vcmpequt expands to wrong
                    pattern
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: wschmidt at gcc dot gnu.org
  Target Milestone: ---

This line appears in a recent patch committed this week:

+BU_P10V_AV_2 (VCMPEQUT,                "vcmpequt",     CONST,  eqvv1ti3)       

This builtin should expand to the ISA 3.1 vcmpequq instruction, but this
definition causes it to expand to bitwise equivalence (not-XOR).  That violates
the definition of vec_cmpeq, which must return all 0s or all 1s for each lane
element.  For this type, that means all 0s or all 1s in the entire register.

[Bug target/101022] rs6000: __builtin_altivec_vcmpequt expands to wrong pattern

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101022

Bill Schmidt <wschmidt at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bergner at gcc dot gnu.org,
                   |                            |carll at gcc dot gnu.org,
                   |                            |segher at gcc dot gnu.org
      Known to fail|                            |12.0
           Keywords|                            |wrong-code
             Status|UNCONFIRMED                 |NEW
             Target|                            |powerpc*-*-*
     Ever confirmed|0                           |1
   Target Milestone|---                         |12.0
   Last reconfirmed|                            |2021-06-10

--- Comment #1 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Confirmed.

[Bug target/101022] rs6000: __builtin_altivec_vcmpequt expands to wrong pattern

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101022

--- Comment #2 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Looks like the proper pattern should be altivec_eqv1ti.

[Bug target/101022] rs6000: __builtin_altivec_vcmpequt expands to wrong pattern

[cel at us dot ibm.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101022

Carl Love <cel at us dot ibm.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |cel at us dot ibm.com

--- Comment #3 from Carl Love <cel at us dot ibm.com> ---
I will work on the fix.

[Bug target/101022] rs6000: __builtin_altivec_vcmpequt expands to wrong pattern

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101022

--- Comment #4 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Hi Carl -- while you're in there, can you please remove these?

+BU_P10_OVERLOAD_2 (VRLQ, "vrlq")                                               
+BU_P10_OVERLOAD_2 (VSLQ, "vslq")                                               
+BU_P10_OVERLOAD_2 (VSRQ, "vsrq")                                               
+BU_P10_OVERLOAD_2 (VSRAQ, "vsraq")                                             

They aren't used in the overload table, because we already have vec_rl, vec_sl,
vec_sr, and vec_sra that you rightly use.  So these macros don't provide any
other utility.

Sorry I didn't see this sooner.

Thanks!
Bill

[Bug target/101022] rs6000: __builtin_altivec_vcmpequt expands to wrong pattern

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101022

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

https://gcc.gnu.org/g:913b13fcb1dceea0e57a04cb77b11097b132cbf6

commit r12-1483-g913b13fcb1dceea0e57a04cb77b11097b132cbf6
Author: Carl Love <cel@us.ibm.com>
Date:   Thu Jun 10 17:36:25 2021 -0500

    Fix for vcmpequt builtin

    The vcmpequt builtin define eqvv1ti3 points to the eqv define instruction
    for the eqv instruction.  The vcmpequt builtin define should point to the
    vector_eqv1ti instruction definition for the vcmpequq instruction.

    2021-06-15  Carl Love  <cel@us.ibm.com>

    gcc/ChangeLog
            PR target/101022
            * config/rs6000/rs6000-builtin.def (VCMPEQUT): Fix the ICODE for
the
            enum definition.
            (VRLQ, VSLQ, VSRQ, VSRAQ): Remove unused BU_P10_OVERLOAD_2
            definitions.

[Bug target/101022] rs6000: __builtin_altivec_vcmpequt expands to wrong pattern

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101022

Bill Schmidt <wschmidt at gcc dot gnu.org> changed:

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

--- Comment #6 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Looks like this was fixed some time ago.