Use a hash of hashes rather than individual scalar count variables. You could use unpack to break the string into words and a global regex match and capture for the individual positions.

knoppix@Microknoppix:~$ perl -Mstrict -MData::Dumper -wE '
> my $str   = q{ATCGGCGCCTAT};
> my @words = unpack q{(a3)*}, $str;
> my %counts;
> 
> foreach my $word ( @words )
> {
>     my $posn;
>     $counts{ q{position } . ++ $posn }->{ $1 } ++
>        while $word =~ m{(.)}g;
> }
> 
> print Data::Dumper->Dumpxs( [ \ %counts ], [ qw{ *counts } ] );'
%counts = (
            'position 1' => {
                              'A' => 1,
                              'T' => 1,
                              'G' => 2
                            },
            'position 3' => {
                              'T' => 1,
                              'C' => 3
                            },
            'position 2' => {
                              'A' => 1,
                              'T' => 1,
                              'C' => 1,
                              'G' => 1
                            }
          );
knoppix@Microknoppix:~$ 
[download]

I hope this is helpful.