Re: Re: Re: Making a match

++ to you and perlplexer. I hadn't realized you could do it that way. Benchmark.pm shows defining the hash is over twice as fast as assigning values to it. The benchmark results on the comparison techniques are more interesting:

if ($hash{$x} == 1)
{
     # always slower than if($hash{$x})
     # or if(defined($hash{$x}))
}

if ($hash{$x})
{
     # slower than if(defined($hash{$x}))
     # when $hash{$x} is undefined

     # faster than if(defined($hash{$x}))
     # when $hash{$x} is defined
}
[download]

Here's the benchmark code I used:

#!/usr/bin/perl

use strict;
use Benchmark;

my @list = (10,11,12,13,14,15,17,18,22,23,24,25,26,27,28,29,30,31,32,3
+3,34,35 );
my %hash;
@hash{@list} = (1) x @list;

timethese(1000000, {
        'Hash Defined Miss' => sub {
                if (defined($hash{1})) {
                        #nop
                }
        },
        'Hash Defined Hit' => sub {
                if (defined($hash{10})) {
                        #nop
                }
        },
        'Hash Assigned Miss Truth' => sub {
                if ($hash{1}) {
                        #nop
                }
        },
        'Hash Assigned Miss Equality' => sub {
                if ($hash{1} == 1) {
                        #nop
                }
        },
        'Hash Assigned Hit Truth' => sub {
                if ($hash{10}) {
                        #nop
                }
        },
        'Hash Assigned Hit Equality' => sub {
                if ($hash{10} == 1) {
                        #nop
                }
        }
});
[download]

Benchmark: timing 1000000 iterations of Hash Assigned Hit Equality, Ha
+sh Assigned Hit Truth,
    Hash Assigned Miss Equality, Hash Assigned Miss Truth, Hash Define
+d Hit, Hash Defined Miss...
Hash Assigned Hit Equality:  5 wallclock secs ( 1.52 usr +  0.00 sys =
+  1.52 CPU) @ 657894.74/s (n=1000000)
Hash Assigned Hit Truth:  3 wallclock secs ( 1.29 usr + -0.01 sys =  1
+.28 CPU) @ 781250.00/s (n=1000000)
Hash Assigned Miss Equality:  3 wallclock secs ( 1.30 usr +  0.00 sys 
+=  1.30 CPU) @ 769230.77/s (n=1000000)
Hash Assigned Miss Truth:  1 wallclock secs ( 1.15 usr +  0.01 sys =  
+1.16 CPU) @ 862068.97/s (n=1000000)
Hash Defined Hit:  1 wallclock secs ( 1.41 usr +  0.00 sys =  1.41 CPU
+) @ 709219.86/s (n=1000000)
Hash Defined Miss:  2 wallclock secs ( 0.95 usr +  0.00 sys =  0.95 CP
+U) @ 1052631.58/s (n=1000000)
[download]

-Matt

Comment on Re: Re: Re: Making a match Select or Download Code

Replies are listed 'Best First'.
Benchmark tips by TheDamian (Vicar) on Apr 14, 2003 at 22:02 UTC
(Sigh. I just can help myself, can I? Always with the advice... ;-) Two suggestions for using Benchmark: If you're using long names for each test, pad the names to the same length with extra spaces on the left. Then your results are much easier to read and compare: Hash Exists Hit: 4 wallclock secs @ 1204819.28/s (n=10000 +00) Hash Defined Hit: 3 wallclock secs @ 2173913.04/s (n=10000 +00) Hash Exists Miss: 1 wallclock secs @ 1515151.52/s (n=10000 +00) Hash Defined Miss: 3 wallclock secs @ 1492537.31/s (n=10000 +00) Hash Assigned Miss Truth: 6 wallclock secs @ 1515151.52/s (n=10000 +00) Hash Assigned Hit Equality: 4 wallclock secs @ 1149425.29/s (n=10000 +00) Hash Assigned Hit Truth: 5 wallclock secs @ 970873.79/s (n=100000 +0) Hash Assigned Miss Equality: 6 wallclock secs @ 877192.98/s (n=100000 +0) [download] Use `cmpthese` instead of `timethese`. That adds a handy comparison matrix at the end of your test results (though this suggestion works much better with shorter names): `Rate HDH HAHE HAME HEH HAMT HDM HAHT HEM HDH 1111111/s -- -0% -4% -21% -24% -28% -30% -39% HAHE 1111111/s 0% -- -4% -21% -24% -28% -30% -39% HAME 1162791/s 5% 5% -- -17% -21% -24% -27% -36% HEH 1408451/s 27% 27% 21% -- -4% -8% -11% -23% HAMT 1470588/s 32% 32% 26% 4% -- -4% -7% -19% HDM 1538462/s 38% 38% 32% 9% 5% -- -3% -15% HAHT 1587302/s 43% 43% 37% 13% 8% 3% -- -13% HEM 1818182/s 64% 64% 56% 29% 24% 18% 15% --` [download]	[reply] [d/l] [select]
Re: Benchmark tips by TheDamian (Vicar) on Apr 14, 2003 at 22:43 UTC
WRT suggestion 1: just thought I should mention that I was, of course, far too Lazy to actually align them myself: `use List::Util 'max'; sub align { my ($hash) = @_; my $len = max map length, keys %$hash; $hash->{sprintf "%${len}s", $_} = delete $hash->{$_} for keys %$hash; return $hash; } cmpthese(10000, align { 'Bacon, Lettuce, & Tomato' => sub {...}, 'Ham on Rye' => sub {...}, 'Spam' => sub {...}, # etc. });` [download]	[reply] [d/l]


XP is just a number
	PerlMonks