use Benchmark 'cmpthese';

my $a = 'foo bar doz' x 100;
$a .= ' hello '.$a;

my $sub = sub {
  /\bhello\b/; /\bhello\b/; /\bhello\b/; /\bhello\b/;
  /\bhello\b/; /\bhello\b/; /\bhello\b/; /\bhello\b/;
};

cmpthese(-3, { loop => sub {
                 for (($a) x 10) {
                   for my $i (1..8) {
                     /\bhello\b/;
                   }
                 }
               },
               sub => sub {
                 for (($a) x 10) {
                   $sub->()
                 }
               },
               inline => sub {
                 for (($a) x 10) {
                   /\bhello\b/; /\bhello\b/; /\bhello\b/; /\bhello\b/;
                   /\bhello\b/; /\bhello\b/; /\bhello\b/; /\bhello\b/;
                 }
               }
[download]

         Rate   loop    sub inline
loop   4157/s     --    -5%   -16%
sub    4363/s     5%     --   -12%
inline 4943/s    19%    13%     --
[download]

open(MAP, "<$new_name_map_file");
while (<MAP>) {
     chomp;
     tr/A-Z/a-z/;
     @map_line = split (/\t/);
     $mapper{$map_line[0]} = $map_line[1];
}
close(MAP);

my $sub = <<'EOS';
sub {
  while (<IN>) {
  print "%";
  tr/A-Z/a-z/;
EOS

for my $name (sort keys %mapper) {
  my $qname = quotemeta $name;
  my $qrepl =  quotemeta $mapper{$name};
  $sub .= "s{\b$qname\b}{$qrepl}g;\n";
}

$sub .= <<'EOS'
    print OUT $_;
  }
}
EOS

$sub = eval $sub;
die if $@;

open(IN, "<input_file");
open(OUT, ">input_file.new");
$sub->();
close(IN);
clse(OUT);
[download]

I guess I am naive. However, I am not sure how your code can speed up things much. I have a wild thought though. Is there any way I can compile Perl code to make it a bin excutable which could be dramatically faster to run...?

I am not sure how your code can speed up things much

well, the bottleneck in the OP code was probably the regular expresion being compiled over and over because he was using a variable ($key) inside. On my code, regexps are just compiled once so it should be much faster.

You could also use the qr operator to precompile them, but expanding the loop should give some additional speed as you can see from the benchmarks.

Is there any way I can compile...

No.

Thanks. Those benchmarks are food for thought.