Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!usc!elroy.jpl.nasa.gov!sdd.hp.com!spool.mu.edu!uunet!convex!usenet
From: tchrist@convex.COM (Tom Christiansen)
Newsgroups: comp.lang.perl
Subject: Re: pattern matching performance
Keywords: grep, pattern matching
Message-ID: <1991May06.193907.3323@convex.com>
Date: 6 May 91 19:39:07 GMT
References: <1991May6.161906.21161@ccu.umanitoba.ca>
Sender: usenet@convex.com (news access account)
Reply-To: tchrist@convex.COM (Tom Christiansen)
Organization: CONVEX Software Development, Richardson, TX
Lines: 69
Nntp-Posting-Host: pixel

From the keyboard of rahardj@ccu.umanitoba.ca (Budi Rahardjo):
:I was wondering if anybody could show me the fastest way to match
:a pattern in perl. I have a big flatfile (around 17000 lines).
:Using UNIX grep takes around 1 or 2 second, but perl's pattern
:matching takes 12 secs. 
:Need advice ...
:
:-- budi
:
:Here is a simplified benchmark that I use 
:
:----- cut here ----
:#!/usr/local/bin/perl
:# Benchmarking pattern matching
:# Sun4 - Sun OS 4.1.1
:#
:print "Enter a pattern to grep : ";
:$pat = <STDIN>; chop $pat;
:
:# Using UNIX grep
:$start = time;
:open (FLAT,"grep $pat flatfile|");
:while (<FLAT>) { print; }
:$elapse = time - $start;
:print ">>>> UNIX grep takes $elapse sec.\n";
:close(FLAT);
:
:# Using perl pattern matching
:open(FLAT,"flatfile");
:$start = time;
:while (<FLAT>) {
:    if (/$pat/) { print;}
:    }
:$elapse = time - $start;
:print ">>>> Perl's pattern matching takes $elapse sec.\n";
:close(FLAT);

First, don't use a block on your if.   It just slows
you down.  Perl has to go through a bit of overhead 
because of your block declaration -- I'm sure Larry
can much better explain than I the magical contortions
he goes through to optimize some of this stuff.  The
short story is that you want to do one of these:

    print if /$pat/;
    /$pat/ && print;

Second, your pattern is invariant, so tell perl this
by adding a /o modifier to your pattern match.

    print if /$pat/o;

Finally, you could change the loop part to an eval so
that you trick the perl compiler into seeing a fixed
expression and thus calling a better checker.  Why Larry
doesn't do this on a /o as well, I do not know, but he
doesn't, and this will run faster than the preceding example:

    open(FLAT,"flatfile");
    $start = time;
    eval "while (<FLAT>) { print if /$pat/; }";
    $elapse = time - $start;
    print ">>>> Perl's pattern matching takes $elapse sec.\n";
    close(FLAT);

--tom
--
Tom Christiansen		tchrist@convex.com	convex!tchrist
		"So much mail, so little time."