Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:82902
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: pass (pb1.pair.com: domain lerdorf.com designates 209.85.192.182 as permitted sender)
Message-ID: <54E27D8D.5020901@lerdorf.com>
Date: Mon, 16 Feb 2015 15:30:21 -0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.4.0
MIME-Version: 1.0
To: Bob Weinand <bobwei9@hotmail.com>, 
 PHP Internals <internals@lists.php.net>
References: <BLU436-SMTP151E12D0D6F709DCC26B14DE22E0@phx.gbl>
In-Reply-To: <BLU436-SMTP151E12D0D6F709DCC26B14DE22E0@phx.gbl>
Content-Type: multipart/signed; micalg=pgp-sha1;
 protocol="application/pgp-signature";
 boundary="qXPXkPJ051Qra3tF11pVmCbiQbKboPukm"
Subject: Re: [PHP-DEV] Switch jumptable optimization
From: rasmus@lerdorf.com (Rasmus Lerdorf)

--qXPXkPJ051Qra3tF11pVmCbiQbKboPukm
Content-Type: text/plain; charset=windows-1252
Content-Transfer-Encoding: quoted-printable

On 02/16/2015 03:04 PM, Bob Weinand wrote:
> I'd like to show you my recent work on a jumptable optimization for swi=
tches.
>=20
> https://github.com/php/php-src/pull/1048 <https://github.com/php/php-sr=
c/pull/1048>
>=20
> It is a fully transparent optimization of switches, by putting a new ZE=
ND_SWITCH opcode at the top of the switch in case we can build a jumptabl=
e, which is the case if the cases are only scalars (no doubles) or evalua=
te to them at compile-time.
>=20
> Switches tend to be big sometimes with a lot of static (literals or con=
stants usually) cases, which was a comparison for each case, and now just=
 a simple hashtable lookup is needed, which greatly improves performance.=
 In case where the result of the switch can be determined at compile-time=
, it even replaces the comparisons with a simple ZEND_JMP opcode.
>=20
> In synthetic benchmarks the results are mind blowing, but in practice, =
benefits mostly are stateful userland parsers (often called with sometime=
s big switches), where it makes it a few percent faster. (For more concre=
te numbers see https://github.com/php/php-src/pull/1048#issuecomment-7303=
2647)
>=20
> As the optimization is only done if there are 3 or more cases, the look=
up is always faster than the single comparisons. So there's no negative p=
erformance impact.
>=20
> It already works, except with opcache where the CFG optimization pass s=
till lacks support. Dmitry or Laruence volunteered to add support for opc=
ache.
>=20
> I'd like to know if there are any technical objections to the PR. If no=
t, they will add opcache support and merge it in.

It looks ok to me if Xinchen or Dmitry gets the opcache support working
for it and the feature is stabilized in time. I gave it a try and it is
segfaulting for me (no opcache):

=3D=3D5612=3D=3D Command: sapi/cli/php -n jump.php
=3D=3D5612=3D=3D
=3D=3D5612=3D=3D Invalid read of size 4
=3D=3D5612=3D=3D    at 0xA64637: _zend_is_inconsistent (zend_hash.c:44)
=3D=3D5612=3D=3D    by 0xA65749: _zend_hash_add_or_update_i (zend_hash.c:=
427)
=3D=3D5612=3D=3D    by 0xA65AF9: _zend_hash_add (zend_hash.c:489)
=3D=3D5612=3D=3D    by 0xA37E28: zend_hash_add_ptr (zend_hash.h:458)
=3D=3D5612=3D=3D    by 0xA39B0A: zend_hash_add_constant (zend_constants.c=
:486)
=3D=3D5612=3D=3D    by 0xA39D01: zend_register_constant (zend_constants.c=
:523)
=3D=3D5612=3D=3D    by 0xA389E9: zend_register_long_constant (zend_consta=
nts.c:194)
=3D=3D5612=3D=3D    by 0xA3855C: zend_register_standard_constants
(zend_constants.c:116)
=3D=3D5612=3D=3D    by 0xA52D6E: zend_startup (zend.c:671)
=3D=3D5612=3D=3D    by 0x9C7FAA: php_module_startup (main.c:2099)
=3D=3D5612=3D=3D    by 0xAFA460: php_cli_startup (php_cli.c:421)
=3D=3D5612=3D=3D    by 0xAFC4D6: main (php_cli.c:1335)
=3D=3D5612=3D=3D  Address 0x8 is not stack'd, malloc'd or (recently) free=
'd

And jump.php is:

<?php
$a =3D 1;
switch($a) {
	case 0: echo 0; break;
	case 1: echo 1; break;
	case 2: echo 1; break;
	case 3: echo 1; break;
	case 4: echo 1; break;
	case 5: echo 1; break;
	case 6: echo 1; break;
	case 7: echo 1; break;
	case 8: echo 1; break;
	case 9: echo 1; break;
}

-Rasmus


--qXPXkPJ051Qra3tF11pVmCbiQbKboPukm
Content-Type: application/pgp-signature; name="signature.asc"
Content-Description: OpenPGP digital signature
Content-Disposition: attachment; filename="signature.asc"

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1

iEYEARECAAYFAlTifY0ACgkQlxayKTuqOuBfnACbBMeArK17jJqEzX2uUDT53B7M
kH0AnRQAaqDPXLtipcoHQZoVfioQAr/H
=t+TN
-----END PGP SIGNATURE-----

--qXPXkPJ051Qra3tF11pVmCbiQbKboPukm--