Use a counter, when you see "loop", ++; when you see "endloop", --. when you see warn, and counter is greater than 0, take it.

use strict;
use warnings;

my $count;
while (my $line = <DATA>) {
    if ($line =~ m/endloop/) {
        $count --;
    } elsif ($line =~ m/loop/) {#don't put this before endloop, as end
+loop itself matches loop
        $count ++;
    } elsif ($line =~ m/warn/) {
        print $line if ($count > 0);
    }
}

__DATA__
begin
  loop
    warn('xxx');
  endloop;

  loop
    null;
      loop
        warn('AAA');
      endloop;
  endloop;
  
  loop
      loop
        warn('BBB');
      endloop;
    warn('rrr');
  endloop;

 begin
   loop
   endloop;

    warn('yyy');
 end;

  warn('zzz');
end;
[download]

You have changed the requirement to include more complicated loop..endloop constructs. While there may be a regex that can dissect that, perhaps a parser is more what you need. Does your data have a grammar you can feed a parser generator like Parse::RecDescent or (b)yacc?

Thanks every one for your help. I had a look at Parse::RecDescent, it looks too advanced for my requirement. I will try to understand it more before deciding on anything. Thanks again for all the help !!

my $loop=0;
my $endloop=0;
while (<>) {
   ++$loop if /\bloop\b/;
   ++$endloop if /\bendloop\b/;
   print if /\bwarn\b/ && $loop > $endloop;
}
[download]

use strict;
my $nest;
for (<DATA>) {
    $nest++ if /\bloop/;
    $nest-- if /\bendloop/;
    $nest && /warn/ && print
}
__DATA__
begin
  loop
    warn('xxx');
  endloop;

  loop
    null;
      loop
        warn('AAA');
      endloop;
  endloop;
  
  loop
      loop
        warn('BBB');
      endloop;
    warn('rrr');
  endloop;

 begin
   loop
   endloop;

    warn('yyy');
 end;

  warn('zzz');
end;
[download]