Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:80658
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: pass (pb1.pair.com: domain zeyos.com designates 109.70.220.166 as permitted sender)
Date: Fri, 16 Jan 2015 19:14:18 +0100
To: Dmitry Stogov <dmitry@zend.com>
Cc: PHP Internals <internals@lists.php.net>,
	Xinchen Hui <xinchen.h@zend.com>
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain;
	charset="UTF-8"
Message-ID: <20150116181421.3870A5F8DD@mx.zeyon.net>
Subject: Re: [PHP-DEV] Generating more efficient code for while-loop
From: ben.coutu@zeyos.com (Benjamin Coutu)

Hi Dimitry,=0A=0AThank you for the acknowledgment. I am a big fan of your w=
ork.=0A=0AI think adding the suggested changes to foreach will be a good im=
provement. With foreach one cold actually merge the jump-into-opcode (line =
0 in the while-loop example) with FE_RESET.=0A=0AOn another note, I think t=
here are many simple compiler optimizations (without the need for real code=
 analysis) that should now be possible, especially in light of AST. The fol=
lowing samples of common patterns come to mind:

if/while (! ...)
BOOL_NOT+JMPZ -> JMPNZ

if/while ( ... &&/|| ... )
BOOL+JMPZ -> JMPZ (BOOL is obsolete)
=0Acase (...)=0ACASE+JMPZ -> CASE (add JMPZ functionality to CASE operation=
)=0A
I understand that these aren't huge improvements per se, but they are very =
common and could at least help reduce the generated bytecode size (thinking=
 OpCache).=0A=0ACheers,=0A=0ABen=0A=0A=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D Origin=
al =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=0AFrom: Dmitry Stogov <dmitry@zend.com>=
=0ATo: Benjamin Coutu <ben.coutu@zeyos.com>=0ADate: Fri, 16 Jan 2015 17:12:=
34 +0100=0ASubject: Re: [PHP-DEV] Generating more efficient code for while-=
loop=0A=0AHi Benjamin,=0A=0AThanks for idea.=0AWith PHP7 AST compiler, it's=
 quite easy to implement this (it took us 15=0Aminutes to try :)=0AHowever,=
 it doesn't make big improvement on our benchmarks.=0AWe will take a look i=
nto possibilities to apply your idea to other patterns=0A(e.g. for and fore=
ach loops).=0A=0AAnyway, it makes full sense to include this optimization i=
nto PHP7.=0AI just like to try other opportunities first.=0A=0AThanks. Dmit=
ry.=0A=0AOn Fri, Jan 16, 2015 at 10:56 AM, Benjamin Coutu <ben.coutu@zeyos.=
com>=0Awrote:=0A=0A> Hello,=0A>=0A> Please consider the following PHP code:=
=0A>=0A> while ( <CONDITION> ) {=0A>   <STATEMENT>=0A> }=0A>=0A> It current=
ly compiles to something like this for the Zend Engine:=0A>=0A> 0: <CONDITI=
ON>=0A> 1: JMPZ -> 4=0A> 2: <STATEMENT>=0A> 3: JMP -> 0=0A> 4: ...=0A>=0A> =
This can easily be rewritten in an equivalent but much more efficient form:=
=0A>=0A> 0: JMP -> 2=0A> 1: <STATEMENT>=0A> 2: <CONDITION>=0A> 3: JMPNZ -> =
1=0A>=0A> The trick is to jump to the conditional expression before the fir=
st=0A> iteration and then only having to continue with just on conditional =
jump on=0A> every subsequent cycle.=0A>=0A> This would result in only 1 jum=
p per loop iteration compared to=0A> effectively 2 jumps needed per iterati=
on with the current implementation.=0A>=0A> It would also make the loop tig=
hter as upon entering the loop (after first=0A> conditional evaluation), on=
e less opcode remains relevant to the loop.=0A>=0A> An analogous approach c=
an be taken for the for-loop.=0A>=0A> Thanks,=0A>=0A> Ben=0A>=0A> PS: This =
is my first post on a PHP mailinglist - I am very excited! :)=0A>=0A> --=0A=
>=0A> Benjamin Coutu=0A> Zeyon Technologies Inc.=0A> http://www.zeyos.com=
=0A>=0A>=0A> --=0A> PHP Internals - PHP Runtime Development Mailing List=0A=
> To unsubscribe, visit: http://www.php.net/unsub.php=0A>=0A>