I rewrote the parser using substr as suggested in the thread you provided. It doesn't seem to have made much of a noticeable difference at all in terms of the speed.

Have I fundamentally misunderstood anything? My implementation is below.

Thanks

# main
#
my $t0 = Benchmark->new();

my $FEED;
($go{file}) ?
    open $FEED, "<:perlio", $go{file} :
    open $FEED, "<&", *STDIN;

binmode $FEED;

my %vtmf_type = (
    0 => \&int_type,
    1 => \&byte_type,
    2 => \&var_length,
    3 => \&long_type,
    4 => sub { return 0 },
    5 => \&subtype_type,
    6 => \&var_length,
    7 => \&price_type,
    __DEFAULT__ => sub { $log->warn("skipping message"); return ('', 2
+55) }
);

my $message_length_bin;
while (read $FEED, $message_length_bin, 2) {
    my $message_length = unpack 'n', $message_length_bin;

    my $offset = 0;
    read $FEED, my($message), $message_length;

    while ($offset < $message_length) {
        my $tag_bin = substr $message, $offset, 2; $offset += 2;
        my $tag_field = unpack 'B16', $tag_bin;
        my $type = oct('0b' . substr $tag_field, 0, 3);
        my $tag = substr $tag_field, 3, 16;

        my $value;
        ($value, $offset) = ($vtmf_type{$type} || $vtmf_type{__DEFAULT
+__})->($message, $offset);

        print oct('0b' . $tag) . '=' . $value . ',';
    }

    print "\n";
}

my $t1 = Benchmark->new();

if ($go{time}) {
    my $td = timediff($t1, $t0);
    $log->info("The code took: ", timestr($td));
}

exit 0;

# subroutines
#
sub int_type {
    my ($message, $offset) = @_;

    my $num_bin = substr $message, $offset,  4; $offset += 4;

    my $value = hex unpack "H*", $num_bin;

    return ($value, $offset);
}

sub long_type {
    my ($message, $offset) = @_;

    my $num_bin = substr $message, $offset, 8; $offset += 8;

    my ($hi, $lo) = map { hex $_ } (unpack 'A8A8', unpack  'H16', $num
+_bin);

    my $num = $hi * 2**32 + $lo;

    return ($num, $offset);
}

sub price_type {
    my ($message, $offset) = @_;

    my $num_bin = substr $message, $offset, 8; $offset += 8;

    my ($hi, $lo) = map { hex $_ } (unpack 'A8A8', unpack  'H16', $num
+_bin);

    my $num = $hi << 8 + $lo;

    $num /= 10**7;

    return ($num, $offset);
}

sub byte_type {
    my ($message, $offset) = @_;

    my $num_bin = substr $message, $offset, 1; $offset += 1;

    my $val = unpack 'A', $num_bin;

    return ($val, $offset);
}

sub subtype_type {
    my ($message, $offset) = @_;

    my $type = substr $message, $offset, 1;

    my ($value, $length) = ('', 0);
    if ($type eq 'B') {
        my $length_bin = substr $message, $offset, 2; $offset += 2;
        my $length = hex unpack 'H4', $length_bin;

        my $value_bin = substr $message, $offset, $length;
        $offset += $length;

    }
    elsif ($type eq 'S') {
        my $length_bin = substr $message, $offset, 2; $offset += 2;
    }
    elsif ($type eq 'P') {
        #skip 2 bytes
        $offset += 2;

        my $num_bin = substr $message, $offset, 8; $offset += 8;
        my ($hi, $lo) = map { hex $_ } (unpack 'A8A8', unpack 'H16', $
+num_bin);

        my $num = $hi << 8 + $lo;

        my $exponent_bin = substr $message, $offset, 2; $offset +=2;
        my $exponent = hex unpack 'H4', $exponent_bin;

        $value = $num * 10**$exponent;
        $length = 13;


        1;
    }
    else {
    }

    return ($value, $length);
}

sub var_length {
    my ($message, $offset) = @_;

    my $length_bin = substr $message, $offset, 1; $offset += 1;
    my $length = hex unpack 'H2', $length_bin;

    my $value_bin = substr $message, $offset, $length; $offset += $len
+gth;
    my $value = unpack "A${length}", $value_bin;

    return ($value, $offset);
}

__END__
[download]

Do you have a sample file?

---

With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday'

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority".

In the absence of evidence, opinion is indistinguishable from prejudice.

Have I fundamentally misunderstood anything

Yes, the three main tricks , but they're not exactly easy to spot unless you're familiar with them :)
1) one single correct unpack call with a clever template is faster than everything else
2) slurping the entire file into one large string is faster than reading byte by byte (its how hard disks work)
3) substr-ing accross a string is faster than chopping a string (or copying a string then chopping it ...
aliasing and/or pass-by-reference is faster than copying

What you did is replace unpack with substr+unpack -- two operations with one -- this will be slower

one of the slow things about your original program is using oct/hex+unpack -- unpack can do most things by itself , see Re: ID3v2 TAG Footer Reading goes wrong (more subs),Re: hex to binary ( UInt32 / Int32 )

reducing the number of calls speeds things up