Re^3: The Boy Scout Rule

in reply to Re^2: The Boy Scout Rule
in thread The Boy Scout Rule

Several problems:

With just 6 comparisons & assignments being run; the overhead of two subroutine calls -- the one you wrapped your tests in and the one Benchmarks wraps those in internally -- becomes a significant factor in the tests.
Pass strings instead of subs to remove one layer of sub-call overhead; and allow benchmark to eval them into subroutines. (It's going to anyway!)
Use an internal loop multiplier to re-balance the test/overhead.
By passing your args wrapped in anonymous arrays -- thus forcing the ternary to do 3 dereferences; and the clever to do 4 dereferences; whereas List::Util only does one -- you bias the tests strongly in List::Util's favour.
Most min() operations will operate on simple scalars so use those instead.
(Minor.) There is little value in using different and big integers; they are all just IVs (or UVs) as far as the comparisons are concerned.
Test for differences in ordering (branch/no branch) by coding separate tests.

The upshot is that the ordering makes no consistent difference (Ie. it flip flops from run to run); and that the ternary is hands down winner for the two simple scalars, common case:

use strictures;
use Benchmark "cmpthese";
use List::Util "min"; # This is XS.

cmpthese -1, {
    list_util_nb   => q[ my( $x, $y ) = ( 0, 1 ); my $m = min( $x, $y 
+)                for 1 .. 1000; ],
    ternary_nb     => q[ my( $x, $y ) = ( 0, 1 ); my $m =      $x < $y
+ ?    $x :  $y   for 1 .. 1000; ],
    clever_nb      => q[ my( $x, $y ) = ( 0, 1 ); my $m = [     $x, $y
+ ]->[ $x <= $y ] for 1 .. 1000; ],
    list_util_b    => q[ my( $x, $y ) = ( 1, 0 ); my $m = min( $x, $y 
+)                for 1 .. 1000; ],
    ternary_b      => q[ my( $x, $y ) = ( 1, 0 ); my $m =      $x < $y
+ ?    $x : $y    for 1 .. 1000; ],
    clever_b       => q[ my( $x, $y ) = ( 1, 0 ); my $m = [    $x, $y 
+]->[  $x <= $y ] for 1 .. 1000; ],
};


__END__
C:\test>junk30
               Rate clever_b clever_nb list_util_nb list_util_b ternar
+y_b ternary_nb
clever_b     1210/s       --      -11%         -67%        -70%      -
+79%       -80%
clever_nb    1356/s      12%        --         -63%        -67%      -
+76%       -78%
list_util_nb 3694/s     205%      172%           --         -9%      -
+34%       -40%
list_util_b  4062/s     236%      200%          10%          --      -
+28%       -33%
ternary_b    5630/s     365%      315%          52%         39%       
+ --        -8%
ternary_nb   6107/s     405%      351%          65%         50%       
+ 8%         --

C:\test>junk30
               Rate clever_nb clever_b list_util_b list_util_nb ternar
+y_b ternary_nb
clever_nb    1297/s        --      -5%        -68%         -69%      -
+75%       -77%
clever_b     1372/s        6%       --        -66%         -67%      -
+74%       -75%
list_util_b  4078/s      214%     197%          --          -3%      -
+22%       -27%
list_util_nb 4190/s      223%     205%          3%           --      -
+20%       -25%
ternary_b    5228/s      303%     281%         28%          25%       
+ --        -6%
ternary_nb   5556/s      328%     305%         36%          33%       
+ 6%         --
[download]

List::Util::min() will obviously win in both speed and clarity for the min( @array ) case.

With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday'

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority". I'm with torvalds on this

In the absence of evidence, opinion is indistinguishable from prejudice. Agile (and TDD) debunked

Comment on Re^3: The Boy Scout Rule Select or Download Code

Replies are listed 'Best First'.
Re^4: The Boy Scout Rule by Your Mother (Archbishop) on Jan 25, 2015 at 21:46 UTC
Thank you for the detailed fix. Unbiasing benchmarks is something I have been needing to improve for a long time.	[reply]

In Section Meditations