Hash references versus hashes: is there a performance hit?

vrk has asked for the wisdom of the Perl Monks concerning the following question:

Honourable Monks,

I often ponder the performance of the code I am writing, but I do this at a relatively high, algorithmical level. I rarely dig into the intricate world of instructions (be it machine language or Perl bytecode). Recently a question has been forming in my mind: is there a performance hit if you use hash references instead of hashes? Allow me to elaborate.

%h = qw( ... );
$h = \%h;

# Accessing the hash and the hash reference in a tight loop.
for ( ... ) {
    my $frobnitz = $h{frobnitz};
    my $quark = $h->{quark};
    ...
}
[download]

Is the latter hash access slower than the former, and if so, how much? Does either consume more memory than the other? Should I be aware of other possible overhead? If I have a complex data structure (HoHoHo...), would I be able to squeeze off some performance from using real hashes instead of hash references?

This is not a burning question, because I very rarely write code in Perl where I would need to mind these things. However, I'm curious.

I know I could possibly test this myself with a profiler, but I have yet to read how to use the Perl profiler, or even to figure out which would be a good test scenario.

Thank you for your patience.

--
print "Just Another Perl Adept\n";

Comment on Hash references versus hashes: is there a performance hit? Download Code

Replies are listed 'Best First'.
Re: Hash references versus hashes: is there a performance hit? by philcrow (Priest) on Mar 06, 2007 at 20:18 UTC
The real difference comes when you pass the hash to a sub. Then, large hashes are serialized as lists, while references are just a single scalar. I'm not willing to guess what the performance of a simple lookup is. But, both of them have the hash the key which might be the dominant part of the process. Phil	[reply]
Re: Hash references versus hashes: is there a performance hit? by jettero (Monsignor) on Mar 06, 2007 at 20:14 UTC
I'm pretty sure those should turn out to be almost exactly the same. If speed is that important, perhaps you should look at assembly language or C or something? You can always import those routines into perl later using XS. You could check using Benchmark: Read more... (369 Bytes) Curiously, `$h{hrmph}` seems to actually be measurably faster, but by a very tiny amount. -Paul	[reply] [d/l] [select]
Re: Hash references versus hashes: is there a performance hit? by chromatic (Archbishop) on Mar 06, 2007 at 20:40 UTC
It doesn't matter. Your program probably spends more time locating and loading modules off of disk than the time difference between accessing a hash by reference and directly.	[reply]
Re: Hash references versus hashes: is there a performance hit? by Anno (Deacon) on Mar 07, 2007 at 00:18 UTC
I know I could possibly test this myself with a profiler, but I have yet to read how to use the Perl profiler, or even to figure out which would be a good test scenario. You don't need a profiler, simple benchmarks will do, as jettero has remarked. The `Benchmark` module is easy to use, in particular its `cmpthese()` function. It is also the only reasonable answer to speed questions because the answer usually depends on the hard- and software environment. That is especially true when the expected difference in performance is small. Anno	[reply] [d/l] [select]