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 795AE3858C54; Sun,  8 Oct 2023 08:45:51 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 795AE3858C54
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1696754751;
	bh=X5qz+hctzrfWiPf0XYvCx2spPzPl6u9IJ6xQdfg2GoA=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=cYr5gj7MCI8IHHz/yJe89R2Kki2BpIBvnsqPuH4JrrHzd5M7cVmcS/QSE2g9kCVvx
	 NbkB+heLTT6hU3gjbnr0XfFZyK3MqXPsaQt6AzAVPA7vIdxCrt7WKtJT7Eio0J5EGO
	 1wD2irEASHrwKyH2VBbx9rVudYEhSqP07u7ZWQK4=
From: "guojie at loongson dot cn" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug target/111403] LoongArch: Wrong code with -O -mlasx
 -fopenmp-simd
Date: Sun, 08 Oct 2023 08:45:46 +0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: target
X-Bugzilla-Version: 14.0
X-Bugzilla-Keywords: wrong-code
X-Bugzilla-Severity: normal
X-Bugzilla-Who: guojie at loongson dot cn
X-Bugzilla-Status: UNCONFIRMED
X-Bugzilla-Resolution: 
X-Bugzilla-Priority: P3
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: ---
X-Bugzilla-Flags: 
X-Bugzilla-Changed-Fields: cc
Message-ID: <bug-111403-4-tOMoUpnwVr@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-111403-4@http.gcc.gnu.org/bugzilla/>
References: <bug-111403-4@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=3D111403

Guo Jie <guojie at loongson dot cn> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |guojie at loongson dot cn

--- Comment #2 from Guo Jie <guojie at loongson dot cn> ---
It seems that =E2=80=9Comp simd reduction=E2=80=9D cannot collaborate well =
with =E2=80=9Cloop peeling=E2=80=9D,
which will result in a probability error in this test case.

LoongArch tree vect pass dump:

  # =E2=80=9Comp simd=E2=80=9D temporary arrays.
  struct S D.3833[8];
  struct S D.3832[8];
  ...


  # prologue loop.
  <bb 20> [local count: 723433550]:
  MEM <struct S[32]> [(struct S *)&D.3832][0].s =3D 0;
  _44 =3D D.3832[0].s;
  _41 =3D (long unsigned int) i_1;
  _58 =3D _41 * 4;
  _59 =3D a_18(D) + _58;
  _60 =3D _59->s;
  _61 =3D _44 + _60;
  D.3832[0].s =3D _61;
  _64 =3D D.3833[0].s;
  _65 =3D D.3832[0].s;
  _66 =3D _64 + _65;
  D.3833[0].s =3D _66;  # Save temporary reduction results.
  MEM <struct S[32]> [(struct S *)&D.3832][0].s =3D _66;
  _69 =3D b_28(D) + _58;
  _70 =3D MEM <struct S[32]> [(const struct S &)&D.3832][0].s;
  _69->s =3D _70;
  i_72 =3D i_1 + 1;
  ivtmp_73 =3D ivtmp_2 - 1;
  ivtmp_78 =3D ivtmp_77 + 1;
  if (ivtmp_78 < prolog_loop_niters.42_7)
    goto <bb 21>; [85.71%]
  else
    goto <bb 18>; [14.29%]
 <bb 21> [local count: 620085901]:
  goto <bb 20>; [100.00%]


  # vector body loop.
  <bb 5> [local count: 118111599]:
  # i_48 =3D PHI <i_30(12), i_79(22)>
  # ivtmp_55 =3D PHI <ivtmp_45(12), ivtmp_81(22)>
  # vectp_a.50_126 =3D PHI <vectp_a.50_127(12), vectp_a.51_123(22)>
  # vectp_b.58_158 =3D PHI <vectp_b.58_159(12), vectp_b.59_155(22)>
  # ivtmp_161 =3D PHI <ivtmp_162(12), 0(22)>
  MEM <vector(8) int> [(struct S *)&D.3832] =3D { 0, 0, 0, 0, 0, 0, 0, 0 };
  _16 =3D (long unsigned int) i_48;
  _17 =3D _16 * 4;
  _19 =3D a_18(D) + _17;
  vect__20.52_128 =3D MEM <vector(8) int> [(int *)vectp_a.50_126];
  _20 =3D _19->s;
  MEM <vector(8) int> [(int *)&D.3832] =3D vect__20.52_128;
  vect__24.54_131 =3D MEM <vector(8) int> [(int *)&D.3833]; # Wrong value.
  ...
  vect__26.56_133 =3D vect__20.52_128 + vect__24.54_131;
  ...
  if (ivtmp_162 < bnd.44_109)
    goto <bb 12>; [0.00%]
  else
    goto <bb 25>; [100.00%]
  ...

The temporary reduction result of =E2=80=9Cprologue loop=E2=80=9D is only s=
tored in D.3833[0],
and all other elements of D.3833 are 0. Therefore, only the first element of
vect__26.56_133 accumulates the scalar reduction result of =E2=80=9Cprologu=
e loop=E2=80=9D.=20

I think the reasonable solution should be to broadcast the scalar reduction
result of =E2=80=9Cprologue loop=E2=80=9D to all elements of D.3833.=