Re: [PHP-DEV] addslashes' speed From: Sascha Schumann (sas <email protected>)
Date: 12/20/98

On Sun, 20 Dec 1998, Zeev Suraski wrote:

> At 04:27 20/12/98 +0100, Sascha Schumann wrote:
> >Hi,
> >
> >the attached patch speeds up addslashes about 30% (benchmarked on 34MB
> >/dev/random data, pgcc 1.1 with -O6 -mcpu=pentium).
> >
> >Btw, using register in a user level program is non sense :-) The effect is
> >rather drastical on a platform with few available registers resulting in
> >nothing but a slowdown.
>
> That's not true. 'register' is a compiler hint telling the compiler that
> this variable is going to be intensively used, and if the code is small
> enough, the compiler should try mapping it to a register. Some compilers
> (noteably, Visual C++) completely ignore this hint, others take it for what
> it is, a hint. I would really hope there's no compiler out there that
> 'takes your word for it' and maps it to a register no matter what. As a
> matter of fact, I'm pretty sure there's no such compiler.

I said the above with PHP in my mind - much code, many variables. If your
compiler takes the register keyword and tries to optimize there, you will
get a slowdown.

Sorry if this statement sounded to general.

> >Anyone objects against a commit?
>
> Other than the part where you remove the register hint, this looks like a
> good optimization. It was pretty dumb on my side to effectively calculate
> that length twice :)

Time for some improvements ... after 3.0.6 is released.

          Regards,
                                                                                
                            Sascha Schumann |
                                 Consultant | finger sas <email protected>
                                            | for PGP public key 

--
PHP Development Mailing List   http://www.php.net/
To unsubscribe send an empty message to php-dev-unsubscribe <email protected>
For help: php-dev-help <email protected>