Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:10430
Return-Path: <valyala@tut.by>
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Delivered-To: mailing list internals@lists.php.net
Received: (qmail 10705 invoked by uid 1010); 14 Jun 2004 08:29:48 -0000
Delivered-To: ezmlm-scan-internals@lists.php.net
Delivered-To: ezmlm-internals@lists.php.net
Received: (qmail 10511 invoked by uid 1007); 14 Jun 2004 08:29:47 -0000
To: internals@lists.php.net
Date: Mon, 14 Jun 2004 11:28:57 +0300
References: <opr9hpiglet8r34v@valyala> <40CC1D3C.9070605@php.net>
Organization: none
Content-Type: text/plain; format=flowed; delsp=yes; charset=koi8-r
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Message-ID: <opr9ksijewt8r34v@valyala>
User-Agent: Opera M2/7.50 (Win32, build 3778)
X-Posted-By: 217.23.116.150
Subject: Re: crc32() improvements
From: valyala@tut.by ("Alexander Valyalkin")

On Sun, 13 Jun 2004 11:24:12 +0200, Ard Biesheuvel <abies@php.net> wrote:

>
> First of all, crc32tab is no longer in the .text segment, so it will not  
> be shared between forked processes, taking more memory space than  
> necessary. Each process will have to initialise it as well, so the init  
> loop will run more than once.
>
> Secondly, compiler optimisation is always more effective working on data  
> known to be constant. So removing 'const' from cr32tab is likely to  
> produce slower code.
>
> Also, you used long int instead of int, which means your code will break  
> on 64-bit platforms.
>
>> 2) Speed is improved on large amount of data.
>
> I seriously doubt that, as the tight main loop of the function didn't  
> change that much. Using 'register' won't help you here (Windows cl.exe  
> officialy ignores it altogether, GCC is smart enough to figure it out  
> for itself)
>

Thank you for good explanation and comments.
Now I understood that the current crc32 implementation is better than mine.
But it consists some ugly bugs (read my comments):
PHP_NAMED_FUNCTION(php_if_crc32)
{
     unsigned int crc = ~0;
     char *p;
     int len, nr; /*!!! is special sense in [len] var here? Remove it! */

     if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "s", &p, &nr) ==  
FAILURE) {
         return;
     }
     /* !!! there is no error check nr < 0 */

     len = 0 ; /* !!! remove it! */
     for (len += nr; nr--; ++p) {
         CRC32(crc, *p);
     }
     RETVAL_LONG(~crc);
}

Below is corrected function with speed improvement in main cycle
(up to 40% on my compiler)
=========cut========
PHP_NAMED_FUNCTION(php_if_crc32)
{
     unsigned int crc = ~0;
     unsigned char *p;
     int nr;
     if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "s", (char **)&p,  
&nr) == FAILURE || nr < 0) return;
     if (nr == 0) RETVAL_LONG(0);
     do { crc = (crc >> 8) ^ crc32tab[(unsigned char)crc ^ *p++]; } while  
(--nr);
     RETVAL_LONG(~crc);
}
=========cut=========

And do not forget to remove CRC32() definition in crc32.h


-- 
Using Opera's revolutionary e-mail client: http://www.opera.com/m2/