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 610E1385ED4B; Tue,  3 Sep 2024 06:08:31 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 610E1385ED4B
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1725343711;
	bh=4Cpgzvtj2LqZ+DN1PdrjPEq1tFXgYHXLRFkwiBb2pTU=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=w1B+U8/ZwX4EcWP4WDZdv0VyyJcVR0rVBhb717eB8LOo8mFMheRAhWjchDeCEGKN1
	 ak8D4BcUwYMjdgqEWEENo2iAsxRTFVH8wKKWMQe3ozVFKulLLgzyYERrYHdBllHUSy
	 OmBuRwMQzYR7fWPtyfsF4cecZy/RdXpWZb1bRD/k=
From: "pinskia at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug rtl-optimization/113682] Branches in branchless binary search
 rather than cmov/csel/csinc
Date: Tue, 03 Sep 2024 06:08:27 +0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: rtl-optimization
X-Bugzilla-Version: 14.0
X-Bugzilla-Keywords: missed-optimization
X-Bugzilla-Severity: normal
X-Bugzilla-Who: pinskia at gcc dot gnu.org
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-113682-4-oMKvrzbtgo@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-113682-4@http.gcc.gnu.org/bugzilla/>
References: <bug-113682-4@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=3D113682

--- Comment #10 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #8)
> This might be the path splitting running on the gimple level causing issu=
es
> too; see PR 112402 .

Actually I was wrong. In this case it is jump threading which messes up the
CFG.

at `-O3 -fno-thread-jumps` on the trunk on aarch64 we get:
        cmp     w0, 0
        csinc   x21, x21, x19, le
        csel    x20, x20, x19, ge

Which is exactly what we want.

The jump threaded IR looks like:
```
  if (_17 > 0)
    goto <bb 4>; [59.00%]
  else
    goto <bb 5>; [41.00%]

  <bb 4> [local count: 598664758]:
  low_19 =3D i_14 + 1;
  goto <bb 6>; [100.00%]

  <bb 5> [local count: 416021267]:
  if (_17 !=3D 0)
    goto <bb 6>; [50.00%]
  else
    goto <bb 8>; [50.00%]

  <bb 6> [local count: 958878294]:
  # low_18 =3D PHI <low_24(5), low_19(4)>
  # high_6 =3D PHI <i_14(5), high_25(4)>
```
Which rtl level ifcvt does not recongize at all as something for cmov.=