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 ACEDA3858D35; Fri,  5 Nov 2021 12:55:10 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org ACEDA3858D35
From: "jakub at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug target/103066]
 __sync_val_compare_and_swap/__sync_bool_compare_and_swap aren't optimized
Date: Fri, 05 Nov 2021 12:55:10 +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: 12.0
X-Bugzilla-Keywords: 
X-Bugzilla-Severity: normal
X-Bugzilla-Who: jakub 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-103066-4-IIlNGvhvrq@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-103066-4@http.gcc.gnu.org/bugzilla/>
References: <bug-103066-4@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
X-BeenThere: gcc-bugs@gcc.gnu.org
X-Mailman-Version: 2.1.29
Precedence: list
List-Id: Gcc-bugs mailing list <gcc-bugs.gcc.gnu.org>
List-Unsubscribe: <https://gcc.gnu.org/mailman/options/gcc-bugs>,
 <mailto:gcc-bugs-request@gcc.gnu.org?subject=unsubscribe>
List-Archive: <https://gcc.gnu.org/pipermail/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-request@gcc.gnu.org?subject=help>
List-Subscribe: <https://gcc.gnu.org/mailman/listinfo/gcc-bugs>,
 <mailto:gcc-bugs-request@gcc.gnu.org?subject=subscribe>
X-List-Received-Date: Fri, 05 Nov 2021 12:55:10 -0000

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=3D103066

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
E.g. the builtin is often used in a loop where the user does his own atomic
load first and decides what to do based on that.
Say for
float f;

void
foo ()
{
  #pragma omp atomic
  f +=3D 3.0f;
}
with -O2 -fopenmp we emit:
  D.2113 =3D &f;
  D.2115 =3D __atomic_load_4 (D.2113, 0);
  D.2114 =3D D.2115;

  <bb 3> :
  D.2112 =3D VIEW_CONVERT_EXPR<float>(D.2114);
  _1 =3D D.2112 + 3.0e+0;
  D.2116 =3D VIEW_CONVERT_EXPR<unsigned int>(_1);
  D.2117 =3D .ATOMIC_COMPARE_EXCHANGE (D.2113, D.2114, D.2116, 4, 0, 0);
  D.2118 =3D REALPART_EXPR <D.2117>;
  D.2119 =3D D.2114;
  D.2114 =3D D.2118;
  if (D.2118 !=3D D.2119)
    goto <bb 3>; [0.00%]
  else
    goto <bb 4>; [100.00%]

  <bb 4> :
  return;
which is essentially
void
foo ()
{
  int x =3D __atomic_load_4 ((int *) &f, __ATOMIC_RELAXED), y;
  float g;
  do
    {
      __builtin_memcpy (&g, &x, 4);
      g +=3D 3.0f;
      __builtin_memcpy (&y, &g, 4);
    }
  while (!__atomic_compare_exchange_n ((int *) &f, &x, y, false,
__ATOMIC_RELAXED, __ATOMIC_RELAXED));
}
Can you explain how your proposed change would improve this?  It would just
slow it down and make it larger.=