I was surprised to see the $& being the fastest. At least on my perl-5.28:

my $string = "AAATTTAGTTCTTAAGGCTGACATCGGTTTACGTCAGCGTTACCCCCCAAGTTATT
+GGGGACTTT";
my %expect = qw( CCCCCC 1 GGGG 1 AAA 1 TTT 3 AA 2 GG 2 TT 5 );

use Test::More;
use Benchmark qw(cmpthese);

my %subs;

sub v1 {
    %subs = ();
    $subs{$_}++ for grep { length >= 2 } split m/,/ => ($string =~ s/(
+[ACGT])\K(?!\1)/,/gr);
    } # v1

sub v2 {
    %subs = ();
    $subs{$_}++ for grep m/^([ACGT])\1+$/ => split m/,/ => ($string =~
+ s/(\w)\K(?!\1)/,/gr);
    } # v2

sub v3 {
    %subs = ();
    $subs{$_}++ for $string =~ m/(AA+|CC+|GG+|TT+)/g;
    } # v3

sub v4 {
    %subs = ();
    $subs{$1}++ while $string =~ m{(([ACGT])\2+)}g;
    } # v4

sub v5 {
    %subs = ();
    $subs{$&}++ while $string =~ m{([ACGT])\1+}g;
    } # v5

v1 (); is_deeply (\%subs, \%expect, "v1");
v2 (); is_deeply (\%subs, \%expect, "v2");
v3 (); is_deeply (\%subs, \%expect, "v3");
v4 (); is_deeply (\%subs, \%expect, "v4");
v5 (); is_deeply (\%subs, \%expect, "v5");

printf "%5d %3d %s\n", $subs{$_->[1]}, @$_ for sort { $b->[0] <=> $a->
+[0] || $a->[1] cmp $b->[1] } map {[ length, $_ ]} keys %subs;

cmpthese (-2, { v1 => \&v1, v2 => \&v2, v3 => \&v3, v4 => \&v4, v5 => 
+\&v5 });

done_testing;
[download]

=>

ok 1 - v1
ok 2 - v2
ok 3 - v3
ok 4 - v4
ok 5 - v5
    1   6 CCCCCC
    1   4 GGGG
    1   3 AAA
    3   3 TTT
    2   2 AA
    2   2 GG
    5   2 TT
      Rate   v2   v1   v3   v4   v5
v2 41981/s   -- -30% -52% -56% -57%
v1 59864/s  43%   -- -31% -38% -39%
v3 87244/s 108%  46%   --  -9% -12%
v4 95919/s 128%  60%  10%   --  -3%
v5 98685/s 135%  65%  13%   3%   --
1..5
[download]