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 75C6D3857803; Thu,  2 Nov 2023 10:48:10 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 75C6D3857803
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1698922090;
	bh=CyreCVkJ0dDJTiBCnogKsNNRW85RrgAbACBWBmIhMTM=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=tMIlC1b2tYo0+I+BPUigfa9zG4Mi/Ayum8zaQUufjfEf6d1x1daCRVfeQKU04BsmH
	 iCjiEgpGdx/BW8HRSbm1CgVVhi8n8kzlfVNWK8rm92FDkR6qlDekcO5TBf7hxgWbRD
	 vANU4kmxzWYuz6cXmSHX9P8w1YJOVELLH/9gWGmM=
From: "rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug tree-optimization/112331] Fail vectorization after loop
 interchange
Date: Thu, 02 Nov 2023 10:48:09 +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: missed-optimization
X-Bugzilla-Severity: normal
X-Bugzilla-Who: rguenth at gcc dot gnu.org
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: 
Message-ID: <bug-112331-4-JcCuAyNiej@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-112331-4@http.gcc.gnu.org/bugzilla/>
References: <bug-112331-4@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=3D112331

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
I'm not sure what the problem is with a zero DR step for an inner loop
reference=20
(possibly dependence analysis runs into some unhandled cases - who knows). =
 The
following vectorizes the inner loop (the load is hoisted as invariant, but
the store is not sunk - there's no sinking phase after interchange).

diff --git a/gcc/tree-vect-data-refs.cc b/gcc/tree-vect-data-refs.cc
index d5c9c4a11c2..7d1f0697fe7 100644
--- a/gcc/tree-vect-data-refs.cc
+++ b/gcc/tree-vect-data-refs.cc
@@ -2944,6 +2944,7 @@ vect_analyze_data_ref_access (vec_info *vinfo,
dr_vec_info *dr_info)
       DR_GROUP_FIRST_ELEMENT (stmt_info) =3D NULL;
       if (!nested_in_vect_loop_p (loop, stmt_info))
        return DR_IS_READ (dr);
+#if 0
       /* Allow references with zero step for outer loops marked
         with pragma omp simd only - it guarantees absence of
         loop-carried dependencies between inner loop iterations.  */
@@ -2954,6 +2955,7 @@ vect_analyze_data_ref_access (vec_info *vinfo,
dr_vec_info *dr_info)
                             "zero step in inner loop of nest\n");
          return false;
        }
+#endif
     }

   if (loop && nested_in_vect_loop_p (loop, stmt_info))


Note when we don't vectorize we are eliding the inner loop later, when
we vectorize we don't.

unvectorized:

s111:
.LFB0:
        .cfi_startproc
        xorl    %eax, %eax
.L2:
        movl    Y(%rax), %ecx
        addq    $4, %rax
        leal    1(%rcx), %edx
        movl    %edx, X-4(%rax)
        cmpq    $128000, %rax
        jne     .L2
        xorl    %eax, %eax
        ret

vectorized:

s111:
.LFB0:
        .cfi_startproc
        movdqa  .LC0(%rip), %xmm1
        xorl    %ecx, %ecx
.L2:
        movdqa  Y(%rcx), %xmm0
        leaq    X(%rcx), %rdx
        movl    $400000, %eax
        paddd   %xmm1, %xmm0
        .p2align 4,,10
        .p2align 3
.L3:
        movaps  %xmm0, (%rdx)
        subl    $2, %eax
        jne     .L3
        addq    $16, %rcx
        cmpq    $128000, %rcx
        jne     .L2
        xorl    %eax, %eax
        ret=