Optimizing sub recompilation

gistfile1.txt

use 5.010;
use strict;
use warnings;

use Benchmark ':all';

my $COUNT = 10_000_000;

say "$COUNT iterations under $^V";

my %totals;
cmpthese( $COUNT, {
    inline => sub {
        runit( sub { ++$totals{inline} } );
    },
    my_sub => sub {
        my $fn = sub { ++$totals{my_sub} };
        runit($fn);
    },
    state_sub => sub {
        state $fn = sub { ++$totals{state_sub} };
        runit($fn);
    },
} );

sub runit {
    $_[0]->();
}

say "All counts in the hash should be $COUNT";
{use Data::Dumper; local $Data::Dumper::Sortkeys=1; say Dumper(\%totals)}


$ perl subs.pl
10000000 iterations under v5.20.3
               Rate    my_sub    inline state_sub
my_sub    1340483/s        --       -8%      -63%
inline    1451379/s        8%        --      -60%
state_sub 3623188/s      170%      150%        --
All counts in the hash should be 10000000
$VAR1 = {
          'inline' => 10000000,
          'my_sub' => 10000000,
          'state_sub' => 10000000
        };