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 EEDA53858D28; Fri, 22 Mar 2024 10:55:49 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org EEDA53858D28
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1711104949;
	bh=cQePxaTOikKbER6SVDxAHgdrtyBdLJ9Kbn87mddQ3zc=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=qAtkwsjLMnEKmn7hdD7euf5y7aghpeZnV+IsWI4cGrAHpk8vJ1VyPGYLr7ZtKX09o
	 HtIoX3gW0RykYr+SuE8jX7E7YqKhc0U084plntaRGtNVD6KoT9uQwhMKp+v3M5/+ui
	 dMv7jqzC6UZW1IusqWHZ5v7K0N2nZFZGe/ahpEa8=
From: "redbeard0531 at gmail dot com" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug tree-optimization/59429] Missed optimization opportunity in
 qsort-style comparison functions
Date: Fri, 22 Mar 2024 10:55:47 +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: 4.8.1
X-Bugzilla-Keywords: missed-optimization
X-Bugzilla-Severity: enhancement
X-Bugzilla-Who: redbeard0531 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-59429-4-2TGys5MbkD@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-59429-4@http.gcc.gnu.org/bugzilla/>
References: <bug-59429-4@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=3D59429

--- Comment #16 from Mathias Stearn <redbeard0531 at gmail dot com> ---
Trunk still generates different code for all cases (in some cases subtly so)
for both aarch64 and x86_64: https://www.godbolt.org/z/1s6sxrMWq. For both
arches, it seems like LE and LG generate the best code.

On aarch64, they probably all have the same throughput, but EL and EG have a
size penalty with an extra instruction.

On x86_64, there is much more variety. EL and EG both get end up with a bra=
nch
rather than being branchless, which is probably bad since comparison functi=
ons
are often called in ways that the result branches are unpredictable. GE and=
 GL
appear to have regressed since this ticket was created. They now do the
comparison twice rather than reusing the flags from the first comparison:

comGL(int, int):
        xor     eax, eax
        cmp     edi, esi
        mov     edx, 1
        setl    al
        neg     eax
        cmp     edi, esi
        cmovg   eax, edx
        ret
comGE(int, int):
        xor     eax, eax
        cmp     edi, esi
        mov     edx, 1
        setne   al
        neg     eax
        cmp     edi, esi
        cmovg   eax, edx
        ret=