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 673733899035; Tue, 15 Nov 2022 16:19:03 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 673733899035
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1668529143;
	bh=r9SajX5UP9HoZP3Mc7jY7daEP1zfdZ8PJvmn9rO+kYA=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=g8952JphPxDGQIK0I9kdadQPiBDzoWNh5zdN0fywARzNvKXmsTc2Pg+1mL0ZKrdzH
	 ynpP+pTCxRs51cJsGzgXeRyMjQyqQ2waDjTVU6Vgc2CECrrZx4CzVcB+ay+0xWjmDZ
	 Mtm4+LPvAyLw1WKQHe43V2jYjdP3czk0Ndx3qTrw=
From: "ibuclaw at gdcproject dot org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug target/107671] i386: Missed optimization: use of bt in bit test
 pattern (using -O2 -mtune=core2)
Date: Tue, 15 Nov 2022 16:19:02 +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: missed-optimization
X-Bugzilla-Severity: normal
X-Bugzilla-Who: ibuclaw at gdcproject dot 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-107671-4-WL5xA5WfkT@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-107671-4@http.gcc.gnu.org/bugzilla/>
References: <bug-107671-4@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=3D107671

--- Comment #5 from Iain Buclaw <ibuclaw at gdcproject dot org> ---
(In reply to Uro=C5=A1 Bizjak from comment #4)
> from:
>         movl    %esi, %ecx
>         movl    $1, %eax
>         sall    %cl, %eax
>         testl   %edi, %eax
>         setne   %al
>         movzbl  %al, %eax
>=20
> to:
>         xorl    %eax, %eax
>         btl     %esi, %edi
>         setc    %al
>=20
> The patch adapts *jcc_bt patterns to similar *scc_bt patterns.
Thanks, have been tinkering with the pointer index version, and found that I
can coax scc_bt_mask to match if I cast the lhs to a signed type. It's not
obvious to me why there would be a difference.

    return ((__INT32_TYPE__)p[bitnum >> 5] & (1 << (bitnum & 31))) !=3D 0;

        movl    %esi, %eax
        shrl    $5, %eax
        movl    (%rdi,%rax,4), %eax
        btl     %esi, %eax
        setc    %al


Things get even stranger once I expand to "bit test and op" variants though
(better to put in another PR though)

    __INT32_TYPE__ btc32(__UINT32_TYPE__ *p, __UINT32_TYPE__ bitnum)
    {
        __INT32_TYPE__ result =3D ((__INT32_TYPE__)p[bitnum >> 5]
                                     & (1 << (bitnum & 31))) !=3D 0;
        p[bitnum >> 5] ^=3D 1 << (bitnum & 31);
        return result;
    }

Patch changes code-gen in the following way.

from:
        movl    %esi, %eax
        movl    %esi, %ecx
        shrl    $5, %eax
        leaq    (%rdi,%rax,4), %rdx
        movl    (%rdx), %eax
        movl    %eax, %esi
        sarl    %cl, %eax
        btcl    %ecx, %esi
        andl    $1, %eax
        movl    %esi, (%rdx)

to:
        movl    %esi, %eax
        shrl    $5, %eax
        leaq    (%rdi,%rax,4), %rdx
        movl    (%rdx), %eax
        movl    %eax, %ecx
        btcl    %esi, %ecx
        btl     %esi, %eax
        setc    %al
        movl    %ecx, (%rdx)
        movzbl  %al, %eax=