From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-bugzilla@gcc.gnu.org>
Received: by sourceware.org (Postfix, from userid 48)
	id B63543858D33; Fri, 14 Apr 2023 07:28:37 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org B63543858D33
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1681457317;
	bh=nPP4dO56NFXv5fFkTvaKFYna2+b5nFqPNbhXCRvqxM8=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=KyaW22Ps0Bl66o96tUU8ACwSNL7mXiVvkQ6ovQVADtovqc6a057vqJrOYymhAwIdb
	 aiBIxnKzb2Ysr++5pDQRGZd+n3Mr8ldLgO+PEQublV2iU+yKcfSOvCnFX0xgnzYx6M
	 /tl33U6tAtPjfmtapRN55a0tXHutQzYd4nnkX8vA=
From: "rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug tree-optimization/109502] [12/13 Regression] wrong code with -O
 -ftree-vectorize -fvect-cost-model=unlimited on aarch64
Date: Fri, 14 Apr 2023 07:28:37 +0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: tree-optimization
X-Bugzilla-Version: 13.0
X-Bugzilla-Keywords: wrong-code
X-Bugzilla-Severity: normal
X-Bugzilla-Who: rguenth at gcc dot gnu.org
X-Bugzilla-Status: NEW
X-Bugzilla-Resolution: 
X-Bugzilla-Priority: P3
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: 12.3
X-Bugzilla-Flags: 
X-Bugzilla-Changed-Fields: 
Message-ID: <bug-109502-4-0lzUQn1pPg@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-109502-4@http.gcc.gnu.org/bugzilla/>
References: <bug-109502-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
List-Id: <gcc-bugs.sourceware.org>

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=3D109502

--- 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:
> >=20
> >   g.0_1 =3D g;
> >   _2 =3D g.0_1 =3D=3D 0;
> >   a_7 =3D (unsigned int) _2;
> >   _3 =3D a_7 % 6;
> >   _4 =3D _3 =3D=3D 0;
> >   _5 =3D (unsigned int) _4;
> >   a_8 =3D _5 + a_7;
> >=20
> > To:
> >=20
> >   g.0_1 =3D g;
> >   _2 =3D g.0_1 =3D=3D 0;
> >   a_7 =3D (unsigned int) _2;
> >   _3 =3D a_7 % 6;
> >   _15 =3D {_3, g.0_1};
> >   mask__4.4_16 =3D { 0, 0 } =3D=3D _15;
> >   vect__5.5_19 =3D VIEW_CONVERT_EXPR<vector(2) unsigned int>(mask__4.4_=
16);
> >   _17 =3D BIT_FIELD_REF <mask__4.4_16, 32, 0>;
> >   _18 =3D (bool) _17;
> >   _4 =3D _3 =3D=3D 0;
> >   _5 =3D (unsigned int) _18;
> >   _20 =3D .REDUC_PLUS (vect__5.5_19);
> >   a_8 =3D _20;
> >=20
>=20
> If anything there is a missing, a negative after the
> reduc_plus (or before) when it translates the bools comparisons into vect=
or
> comparisons.

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

unsigned foo (unsigned *p)
{
  unsigned tem1 =3D p[0] =3D=3D 0;
  unsigned tem2 =3D p[1] =3D=3D 0;
  unsigned tem3 =3D p[2] =3D=3D 0;
  unsigned tem4 =3D p[3] =3D=3D 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:   =3D=3D> examining statement: tem4_16 =3D (unsigned int) _=
8;
t.c:7:29: note:   vect_is_simple_use: operand _7 =3D=3D 0, type of def: int=
ernal
t.c:7:29: missed:   type conversion to/from bit-precision unsupported.
t.c:7:29: note:   vect_is_simple_use: operand _7 =3D=3D 0, type of def: int=
ernal
t.c:7:29: missed:   mixed mask and nonmask vector types=