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 03AE43858C56; Thu, 13 Oct 2022 12:05:31 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 03AE43858C56
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1665662732;
	bh=dClpybrTx5cCGGiW09mKdbT25ZshlI9A4CiA++XCDUE=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=jv62uBWKXLYcroSJp5jaIRG56ZvEO2qLrZapNXucVJSQ9MEBl8XmW+ACQNTeot2aF
	 aGA7Nb0pTm5BOkxDItXBQwKcFrKa6OmQ1gBrni20BXj8NkpWhveoilZDOY6/XuQ45h
	 XAfPY5V5OKs0jgigtQwlPeEC0w1Q+5LhcE0N3a3k=
From: "rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug target/107160] [13 regression] r13-2641-g0ee1548d96884d  causes
 verification failure in spec2006
Date: Thu, 13 Oct 2022 12:05:31 +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: 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: linkw at gcc dot gnu.org
X-Bugzilla-Target-Milestone: 13.0
X-Bugzilla-Flags: 
X-Bugzilla-Changed-Fields: 
Message-ID: <bug-107160-4-TxSh4BD3wA@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-107160-4@http.gcc.gnu.org/bugzilla/>
References: <bug-107160-4@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=3D107160

--- Comment #14 from Richard Biener <rguenth at gcc dot gnu.org> ---
Aha, so the issue is that we have a vectorized epilogue here and the epilog=
ue
of _that_ ends up doing


  <bb 11> [local count: 94607391]:
  # sum0_48 =3D PHI <sum0_28(3)>
  # sum1_47 =3D PHI <sum1_29(3)>
  # sum2_46 =3D PHI <sum2_30(3)>
  # sum3_45 =3D PHI <sum3_31(3)>
  # vect_sum3_31.16_101 =3D PHI <vect_sum3_31.16_97(3)>
  # vect_sum3_31.16_102 =3D PHI <vect_sum3_31.16_98(3)>
  # vect_sum3_31.16_103 =3D PHI <vect_sum3_31.16_99(3)>
  # vect_sum3_31.16_104 =3D PHI <vect_sum3_31.16_100(3)>
  _105 =3D BIT_FIELD_REF <vect_sum3_31.16_101, 32, 0>;
  _106 =3D BIT_FIELD_REF <vect_sum3_31.16_101, 32, 32>;
...
this is from the main vect

  <bb 17> [local count: 81467476]:
  # sum0_135 =3D PHI <sum0_62(12)>
  # sum1_136 =3D PHI <sum1_58(12)>
  # sum2_137 =3D PHI <sum2_54(12)>
  # sum3_138 =3D PHI <sum3_50(12)>
  # vect_sum3_50.25_159 =3D PHI <vect_sum3_50.25_158(12)>
this is from the epilogue

  <bb 15> [local count: 105119324]:
  # sum0_15 =3D PHI <sum0_135(17), _129(11)>
  # sum1_77 =3D PHI <sum1_136(17), _130(11)>
  # sum2_75 =3D PHI <sum2_137(17), _131(11)>
  # sum3_13 =3D PHI <sum3_138(17), _132(11)>
  # _160 =3D PHI <vect_sum3_50.25_159(17), vect_sum3_31.16_101(11)>
  _161 =3D BIT_FIELD_REF <_160, 32, 0>;
  _162 =3D BIT_FIELD_REF <_160, 32, 32>;
  _163 =3D BIT_FIELD_REF <_160, 32, 64>;
  _164 =3D BIT_FIELD_REF <_160, 32, 96>;
  _74 =3D _161 + _162;
  _76 =3D _74 + _163;
  _78 =3D _76 + _164;

so we fail to accumulate the main loops accumulators and just
use the first one.  On x86 the vectorized epilogue uses a smaller
vector size but the same number of accumulators.

It seems it's simply unexpected to have the unrolled SLP reduction
and a vectorized epilogue with the same vector mode (but not unrolled).

I can reproduce the failure when patching the x86 cost model to force
unrolling by 2 (maybe we want a --param to force that to aid debugging...).=