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 D9BF63858D28; Wed, 21 Jun 2023 09:46:52 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org D9BF63858D28
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1687340812;
	bh=zXveg5xPZiDCXnp7wlPCB0UX89T6hJBI1cR0o3H4I3w=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=oGH2a/6ST3HrlX6ZanNuAnIqsXb8hvDGtmBr+z4m5hN+t01+S4wc4uMfIcAyWDJ+l
	 54uz4/cOVibIwXDVVHkPX+a+eJ8Xg88RtbWGXYI05WS3DZNBKv+cYQRbNhRtPo4aOn
	 0GCzsOeq6ttlbsm5SiCV6SHf/wnbm3jG6Y36ffVo=
From: "ubizjak at gmail dot com" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug target/109812] GraphicsMagick resize is a lot slower in GCC
 13.1 vs Clang 16 on Intel Raptor Lake
Date: Wed, 21 Jun 2023 09:46:43 +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: 13.1.1
X-Bugzilla-Keywords: missed-optimization
X-Bugzilla-Severity: normal
X-Bugzilla-Who: ubizjak at gmail dot com
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: ---
X-Bugzilla-Flags: 
X-Bugzilla-Changed-Fields: 
Message-ID: <bug-109812-4-Aw3nkKmkHt@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-109812-4@http.gcc.gnu.org/bugzilla/>
References: <bug-109812-4@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=3D109812

--- Comment #18 from Uro=C5=A1 Bizjak <ubizjak at gmail dot com> ---
One interesting observation:

clang is able to do this:

  0.09 =E2=94=82     =E2=94=82  vmovddup     -0x8(%rdx,%rsi,1),%xmm3       =
       =E2=96=92
  ...
  0.11 =E2=94=82     =E2=94=82  vfmadd231sd  %xmm2,%xmm3,%xmm1             =
       =E2=96=92
  ...
  0.74 =E2=94=82     =E2=94=82  vfmadd231pd  %xmm2,%xmm3,%xmm0             =
       =E2=96=92

It figures out that duplicated V2DFmode value in %xmm3 can also be accessed=
 in
the same register as DFmode value.

OTOH, current gcc does:

        vmovsd  (%rsi,%rax,8), %xmm1
        ...
        vmovddup        %xmm1, %xmm4
        ...
        vfmadd231pd     %xmm4, %xmm0, %xmm2
        ...
        vfmadd231sd     %xmm1, %xmm0, %xmm3

The above code needs two registers.=