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 1264B3858425; Tue,  3 Oct 2023 12:19:12 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 1264B3858425
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1696335552;
	bh=A+YXB2aBWInFeB1IVpj3eOf99TCyRUXGGbr+MLA2JIE=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=byopCLDexWtnMLH94iBaoi2zF5vq95Q+D1RMfsvQ4sMLGoHqKBDBGYFb2NTIjPjo7
	 /95yfxnRL0AZScKRaGAyoacowSC/I9j4wZYS+KdS05kwLu01F4MBSABPLAaOUjMeev
	 gGY8zn/XHbTi+6qvBbxuAD2oVWMp/lNuPzXYW3Hk=
From: "prathamesh3492 at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug tree-optimization/111648] [14 Regression] Wrong code at -O2/3
 on x86_64-linux-gnu since r14-3243-ga7dba4a1c05
Date: Tue, 03 Oct 2023 12:19:11 +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: 14.0
X-Bugzilla-Keywords: wrong-code
X-Bugzilla-Severity: normal
X-Bugzilla-Who: prathamesh3492 at gcc dot gnu.org
X-Bugzilla-Status: ASSIGNED
X-Bugzilla-Resolution: 
X-Bugzilla-Priority: P3
X-Bugzilla-Assigned-To: prathamesh3492 at gcc dot gnu.org
X-Bugzilla-Target-Milestone: 14.0
X-Bugzilla-Flags: 
X-Bugzilla-Changed-Fields: attachments.created
Message-ID: <bug-111648-4-htJmlUI4CQ@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-111648-4@http.gcc.gnu.org/bugzilla/>
References: <bug-111648-4@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=3D111648

--- Comment #3 from prathamesh3492 at gcc dot gnu.org ---
Created attachment 56037
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=3D56037&action=3Dedit
Untested fix

The issue is that when a1 is a multiple of vector length, we end up creating
following encoding in result: { base_elem, arg[0], arg[1], ... } where arg =
is
chosen input vector, which is incorrect.

For above case, vectorizer pass creates VEC_PERM_EXPR<arg0, arg, sel> where:
arg0: { -16, -9, -10, -11 }=20
arg1: { -12, -5, -6, -7 }=20
sel =3D { 3, 4, 5, 6 }

arg0, arg1 and sel are encoded with npatterns =3D 1 and nelts_per_pattern =
=3D 3.
Since a1 =3D 4 and arg_len =3D 4, it ended up creating the result with
following encoding:
res =3D { arg0[3], arg1[0], arg1[1] } // npatterns =3D 1, nelts_per_pattern=
 =3D 3
    =3D { -11, -12, -5 }

So for res[4], it used S =3D (-5) - (-12) =3D 7
And hence computed it as -5 + 7 =3D 2.
instead of arg1[2], ie, -6.
which is the difference we see in output at -O0 vs -O2.

The patch tweaks the constratints in valid_mask_for_fold_vec_perm_cst_p to =
punt
if a1 is a multiple of vector length, so a1 ... ae only selects from stepped
part of the input vector, which seems to fix this issue.
I will run a proper bootstrap+test and post it upstream.=