Perl, cmd, $ARGV[0], slow - performance

Question

[Strawberry Perl v5.16.3, Windows 7 x64, executing via cmd, eg c:\strawberry> perl test.pl 100000]
SYMPTOM: The following code: foreach (1..$ARGV[0]) { foo($_); }, executes roughly 20% slower than if I had included this extra line, before it: my $num = $ARGV[0];
QUESTION: Can anyone help me understand why?
Notice, in the second case, that after I initialize and set $num, I do not then use $num in the loop parameters. Were this the case, I could probably be convinced that repeatedly testing against $ARGV[0] in a forloop is somehow slower than a variable that I define myself... but this is not the case.
To track time, I use: use Time::HiRes; my $time = [Time::HiRes::gettimeofday()]; at the top of my script, and: print "\n1: ", Time::HiRes::tv_interval($time); at the bottom.
Confused!
Thanks,
Michael
EDIT
I am including the entire script, with a comment preceding the offending line... Interestingly, it looks like the time discrepancy is at least partially dependent on my redundant initialization of %h, as well as #chain... This is getting weird.
use Time::HiRes; my $time = [Time::HiRes::gettimeofday()];
#my $max=$ARGV[0];
my %h = (1=>1,89=>89);
$h{1}=1;
$h{89}=89;
my #chain=();
my $ans=0;
sub sum{my $o=0; foreach (#_){$o+=$_}; return $o;}
foreach (1..$ARGV[0]-1){
my $x=$_;
my #chain = ();
while(!exists($h{$x})){
push(#chain,$x);
$x = sum(map {$_**2} split('',$x));
}
foreach (#chain){$h{$_}=$h{$x} if !exists($h{$_});}
}
print "\n1: ", Time::HiRes::tv_interval($time);
foreach (1..$ARGV[0]){$ans++ if ($h{$_}==89);}
print "\n2: ", Time::HiRes::tv_interval($time);

Answer 1

On my system (perl 5.16.3 on GNU/Linux) there is no measurable difference. The standard deviation of the timings is larger than the difference between measurements of different versions.
For each variant of the script, 10 executions were performed. The $ARGV[0] was 3.5E5 in all cases (350000).
Without my $num = $ARGV[0]:
$ perl measure.pl
2.369921 2.38991 2.380969 4.419895 2.398861 2.420928 2.388721 2.368144 2.387212 2.386347
mean: 2.5910908
sigma: 0.609763793801797
With my $num = $ARGV[0]:
$ perl measure.pl
4.435764 2.419485 2.403696 2.401771 2.411345 2.466776 4.408127 2.416889 2.389191 2.397409
mean: 2.8150453
sigma: 0.803721101668365
The measure.pl script:
use strict; use warnings; use 5.016;
use List::Util 'sum';
my #times = map qx/perl your-algorithm.pl 3.5E5/, 1..10;
chomp #times;
say "#times";
say "mean: ", mean(#times);
say "sigma: ", sigma(#times);
sub mean { sum(#_)/#_ }
sub sigma {
my $mean = mean(#_);
my $variance = sum(map { ($_-$mean)**2 } #_) / #_;
sqrt $variance;
}
With your-algorithm.pl being reduced so that only one timing is printed:
foreach (1..$ARGV[0]){$ans++ if ($h{$_}==89);}
print Time::HiRes::tv_interval($time), "\n";