[Bug target/114783] New: [14 Regression] Equality compares of vector builtins spill one operand to the stack

[Bug target/114783] New: [14 Regression] Equality compares of vector builtins spill one operand to the stack

[mkretz at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114783

            Bug ID: 114783
           Summary: [14 Regression] Equality compares of vector builtins
                    spill one operand to the stack
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: mkretz at gcc dot gnu.org
  Target Milestone: ---
            Target: x86_64-*-*, i?86-*-*

Testcase:

https://compiler-explorer.com/z/Y1aTj7Mv9

using V [[gnu::vector_size(32)]] = int;

auto f(V x, V y)
{
  return x == y;
}

auto h(V x, V y)
{
  return x <= y;
}

Compile with -O2 -march=skylake.

[Bug target/114783] [14 Regression] Equality compares of vector builtins spill one operand to the stack

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114783

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2024-04-19
                 CC|                            |jakub at gcc dot gnu.org
   Target Milestone|---                         |14.0
     Ever confirmed|0                           |1
           Priority|P3                          |P1

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r14-4459-gd77ee4a7f7a80596d91b53c06946ac215614e6c4 or so
( r14-4453-gccdc0f0fcf6f240d5c4f059e6f38547a0cca9723 still emitted the 13 code,
and
I don't have the revisions in between in the seed, maybe they failed to
build?).

[Bug target/114783] [14 Regression] Equality compares of vector builtins spill one operand to the stack

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114783

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
@@ -16831,7 +16882,7 @@ (define_insn "*avx2_eq<mode>3"
   [(set (match_operand:VI_256 0 "register_operand" "=x")
        (eq:VI_256
          (match_operand:VI_256 1 "nonimmediate_operand" "%x")
-         (match_operand:VI_256 2 "nonimmediate_operand" "xm")))]
+         (match_operand:VI_256 2 "nonimmediate_operand" "jm")))]
   "TARGET_AVX2 && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
   "vpcmpeq<ssemodesuffix>\t{%2, %1, %0|%0, %1, %2}"
   [(set_attr "type" "ssecmp")

That looks like it was supposed to be xjm.

[Bug target/114783] [14 Regression] Equality compares of vector builtins spill one operand to the stack

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114783

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

[Bug target/114783] [14 Regression] Equality compares of vector builtins spill one operand to the stack

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114783

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 57991
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57991&action=edit
gcc14-pr114783.patch

Full untested patch.

[Bug target/114783] [14 Regression] Equality compares of vector builtins spill one operand to the stack

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114783

--- Comment #4 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:c23db3ebb65ba357155be85ef56d037403eaee36

commit r14-10047-gc23db3ebb65ba357155be85ef56d037403eaee36
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Sat Apr 20 00:13:49 2024 +0200

    i386: Fix up *avx2_eq<mode>3 constraints [PR114783]

    The r14-4456 change (part of APX EGPR support) seems to have mistakenly
    changed in the
    @@ -16831,7 +16831,7 @@ (define_insn "*avx2_eq<mode>3"
       [(set (match_operand:VI_256 0 "register_operand" "=x")
            (eq:VI_256
              (match_operand:VI_256 1 "nonimmediate_operand" "%x")
    -         (match_operand:VI_256 2 "nonimmediate_operand" "xm")))]
    +         (match_operand:VI_256 2 "nonimmediate_operand" "jm")))]
       "TARGET_AVX2 && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
       "vpcmpeq<ssemodesuffix>\t{%2, %1, %0|%0, %1, %2}"
       [(set_attr "type" "ssecmp")
    hunk the xm constraint to jm, while in many other spots it changed
correctly
    xm to xjm.  The instruction doesn't require the last operand to be in
    memory, it can handle 3 256-bit registers just fine, just it is a VEX only
    encoded instruction and so can't allow APX EGPR regs in the memory operand.

    The following patch fixes it, so that we don't force one of the == operands
    into memory all the time.

    2024-04-20  Jakub Jelinek  <jakub@redhat.com>

            PR target/114783
            * config/i386/sse.md (*avx2_eq<mode>3): Change last operand's
            constraint from "jm" to "xjm".

            * gcc.target/i386/avx2-pr114783.c: New test.

[Bug target/114783] [14 Regression] Equality compares of vector builtins spill one operand to the stack

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114783

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed.