my $d = [[],[], [], []];
[download]

You don't need to pre-initialise your deep data structures. They'll pop up automagically when you first use them. It's called autovivification. So a

my $d;

should suffice.

Update: To avoid repetitive pushes you could define this function:

sub pushsome {
  my $AoAref = shift;
  for (my $i = 0 ; $i <= $#_ ; $i++) {
    push @{$AoAref->[$i]}, $_[$i];
  }
}
[download]

and then

while ( <DATA> )
{
  chomp;
  pushsome $d, split /,/;
}
[download]

However in this case you need to pre-define $d as an array reference (my $d = [];) ; first-level autovivification can't work through subroutine calls.

Better off IMO is to get rid of the "reference to an array of arrays" ($r) and use an up-front @AoA. Here's your revamped code:

#!/usr/bin/perl

use strict;
use warnings;

use PDL;
use PDL::Matrix;

sub pushsome {
  my $AoAref = shift;
  for (my $i = 0 ; $i <= $#_ ; $i++) {
    push @{$AoAref->[$i]}, $_[$i];
  }
}

sub fetchdata {
  my $file = shift;
  my @AoA;

  open(DATA, $file) || die "Error: Unable to open $file:$!\n";

  while (<DATA>) {
    chomp;
    pushsome \@AoA, split /,/;
  }

  close DATA;

  PDL::Matrix->pdl(\@AoA);
}


# main program

@ARGV or die "Please supply a data file on the command line!\n";

my $d = fetchdata( shift );

# ...
[download]