An alternative to doing the whole thing with a regex would be to split into individual path elements then push joined elements onto the array, popping elements off the end until there's nothing left.

use 5.026;
use warnings;

use Data::Dumper;

my @paths = qw{
   /abc/def/ghi
   /wxy/z
   /usr/local/lib/x86_64-linux-gnu/perl/5.26.1
   bin/fred
   somefile
   };

foreach my $path ( @paths )
{
    my @elems = split m{/}, $path;
    my @arr;
    while ( @elems )
    {
        last if @elems == 1 && ! $elems[ 0 ]; # Ignore empty first ele
+ment
                                              # if path starts with a 
+/
        push @arr, join q{/}, @elems;
        pop @elems;
    }
    say $path;
    print Data::Dumper->Dumpxs( [ \ @arr ], [ qw{ *arr } ] );
    say q{-} x 30;
}
[download]

The output.

/abc/def/ghi
@arr = (
         '/abc/def/ghi',
         '/abc/def',
         '/abc'
       );
------------------------------
/wxy/z
@arr = (
         '/wxy/z',
         '/wxy'
       );
------------------------------
/usr/local/lib/x86_64-linux-gnu/perl/5.26.1
@arr = (
         '/usr/local/lib/x86_64-linux-gnu/perl/5.26.1',
         '/usr/local/lib/x86_64-linux-gnu/perl',
         '/usr/local/lib/x86_64-linux-gnu',
         '/usr/local/lib',
         '/usr/local',
         '/usr'
       );
------------------------------
bin/fred
@arr = (
         'bin/fred',
         'bin'
       );
------------------------------
somefile
@arr = (
         'somefile'
       );
------------------------------
[download]

A little more long-winded but possibly simpler to understand. I hope this is helpful.