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 E93223858D3C; Fri, 16 Jun 2023 15:37:01 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org E93223858D3C
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1686929821;
	bh=LUqSL50l+r2vsFiyfg1lBbXBIhgCnacmi1NLbOx5wOE=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=CJzwzLMjtFrT+W9VHOZOY3jRFI9W6wtSsO5PFDlwLKro8vShOTwezr9Dw2fdvrqWs
	 h5KpgB18PXsXGQ4PVcoQUETTtuLt0MFLX3+GJ6MBIbIZOk9/O3zTiWwSWTBZT7P8L7
	 zSAGcguPryJ1PdLGyXfQ3ITR5Maahcb6rnFqQKbQ=
From: "hubicka at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug libstdc++/110287] _M_check_len is expensive
Date: Fri, 16 Jun 2023 15:37:01 +0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: libstdc++
X-Bugzilla-Version: 13.1.0
X-Bugzilla-Keywords: 
X-Bugzilla-Severity: normal
X-Bugzilla-Who: hubicka at gcc dot gnu.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-110287-4-emjEFXYBPg@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-110287-4@http.gcc.gnu.org/bugzilla/>
References: <bug-110287-4@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=3D110287

--- Comment #2 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
With patch in PR110289 to optimize the std::max int MAX_EXPR and the throw
commented out I get:

size_type std::vector<std::pair<unsigned int, unsigned int> >::_M_check_len
(const struct vector * const this, size_type __n, const char * __s)
{
  const size_type __len;
  size_type iftmp.2_1;
  long unsigned int _2;
  long unsigned int _5;
  long unsigned int _7;
  struct pair * _8;
  struct pair * _9;
  long int _10;
  long int _11;
  long unsigned int _12;

  <bb 2> [local count: 1073741824]:
  _8 =3D this_4(D)->D.26707._M_impl.D.26014._M_finish;
  _9 =3D this_4(D)->D.26707._M_impl.D.26014._M_start;
  _10 =3D _8 - _9;
  _11 =3D _10 /[ex] 8;
  _12 =3D (long unsigned int) _11;
  _7 =3D __n;
  _5 =3D MAX_EXPR <_7, _12>;
  __len_6 =3D _5 + _12;
  if (__len_6 < _12)
    goto <bb 4>; [35.00%]
  else
    goto <bb 3>; [65.00%]

  <bb 3> [local count: 697932185]:
  _2 =3D MIN_EXPR <__len_6, 1152921504606846975>;

  <bb 4> [local count: 1073741824]:
  # iftmp.2_1 =3D PHI <1152921504606846975(2), _2(3)>
  return iftmp.2_1;

}

which fits early inlining limits. So modifying libstdc++ to avoid the throw
would be great.

Again optimized _M_check_len could avoid:
  <bb 3> [local count: 697932185]:
  _2 =3D MIN_EXPR <__len_6, 1152921504606846975>;

  <bb 4> [local count: 1073741824]:
  # iftmp.2_1 =3D PHI <1152921504606846975(2), _2(3)>
  return iftmp.2_1;

since __len can not get that large without running out of memory. Which wou=
ld
help further inlining.

Testcase from PR109849 now runs in 0.39s while without these two changes it
runs in 0.528s, clang's build runs in 0.057s, so still a long way to go.=